Computer Vision and Pattern Recognition (cs.CV)

Abstract: Manipulatives used in the right way help improve mathematical concepts leading to better learning outcomes. In this paper, we present a phygital (physical + digital) curriculum inspired teaching system for kids aged 5-8 to learn geometry using shape tile manipulatives. Combining smaller shapes to form larger ones is an important skill kids learn early on which requires shape tiles to be placed close to each other in the play area. This introduces a challenge of oriented object detection for densely packed objects with arbitrary orientations. Leveraging simulated data for neural network training and light-weight mobile architectures, we enable our system to understand user interactions and provide real-time audiovisual feedback. Experimental results show that our network runs real-time with high precision/recall on consumer devices, thereby providing a consistent and enjoyable learning experience.

Abstract: Transformer-based image denoising methods have achieved encouraging results in the past year. However, it must uses linear operations to model long-range dependencies, which greatly increases model inference time and consumes GPU storage space. Compared with convolutional neural network-based methods, current Transformer-based image denoising methods cannot achieve a balance between performance improvement and resource consumption. In this paper, we propose an Efficient Wavelet Transformer (EWT) for image denoising. Specifically, we use Discrete Wavelet Transform (DWT) and Inverse Wavelet Transform (IWT) for downsampling and upsampling, respectively. This method can fully preserve the image features while reducing the image resolution, thereby greatly reducing the device resource consumption of the Transformer model. Furthermore, we propose a novel Dual-stream Feature Extraction Block (DFEB) to extract image features at different levels, which can further reduce model inference time and GPU memory usage. Experiments show that our method speeds up the original Transformer by more than 80%, reduces GPU memory usage by more than 60%, and achieves excellent denoising results. All code will be public.

Abstract: Despite the success of multimodal learning in cross-modal retrieval task, the remarkable progress relies on the correct correspondence among multimedia data. However, collecting such ideal data is expensive and time-consuming. In practice, most widely used datasets are harvested from the Internet and inevitably contain mismatched pairs. Training on such noisy correspondence datasets causes performance degradation because the cross-modal retrieval methods can wrongly enforce the mismatched data to be similar. To tackle this problem, we propose a Meta Similarity Correction Network (MSCN) to provide reliable similarity scores. We view a binary classification task as the meta-process that encourages the MSCN to learn discrimination from positive and negative meta-data. To further alleviate the influence of noise, we design an effective data purification strategy using meta-data as prior knowledge to remove the noisy samples. Extensive experiments are conducted to demonstrate the strengths of our method in both synthetic and real-world noises, including Flickr30K, MS-COCO, and Conceptual Captions.

Abstract: We present NeRFVS, a novel neural radiance fields (NeRF) based method to enable free navigation in a room. NeRF achieves impressive performance in rendering images for novel views similar to the input views while suffering for novel views that are significantly different from the training views. To address this issue, we utilize the holistic priors, including pseudo depth maps and view coverage information, from neural reconstruction to guide the learning of implicit neural representations of 3D indoor scenes. Concretely, an off-the-shelf neural reconstruction method is leveraged to generate a geometry scaffold. Then, two loss functions based on the holistic priors are proposed to improve the learning of NeRF: 1) A robust depth loss that can tolerate the error of the pseudo depth map to guide the geometry learning of NeRF; 2) A variance loss to regularize the variance of implicit neural representations to reduce the geometry and color ambiguity in the learning procedure. These two loss functions are modulated during NeRF optimization according to the view coverage information to reduce the negative influence brought by the view coverage imbalance. Extensive results demonstrate that our NeRFVS outperforms state-of-the-art view synthesis methods quantitatively and qualitatively on indoor scenes, achieving high-fidelity free navigation results.

Abstract: We propose a novel Text-to-Image Generation Network, Adaptive Layout Refinement Generative Adversarial Network (ALR-GAN), to adaptively refine the layout of synthesized images without any auxiliary information. The ALR-GAN includes an Adaptive Layout Refinement (ALR) module and a Layout Visual Refinement (LVR) loss. The ALR module aligns the layout structure (which refers to locations of objects and background) of a synthesized image with that of its corresponding real image. In ALR module, we proposed an Adaptive Layout Refinement (ALR) loss to balance the matching of hard and easy features, for more efficient layout structure matching. Based on the refined layout structure, the LVR loss further refines the visual representation within the layout area. Experimental results on two widely-used datasets show that ALR-GAN performs competitively at the Text-to-Image generation task.

Abstract: This paper proposes a novel module called middle spectrum grouped convolution (MSGC) for efficient deep convolutional neural networks (DCNNs) with the mechanism of grouped convolution. It explores the broad "middle spectrum" area between channel pruning and conventional grouped convolution. Compared with channel pruning, MSGC can retain most of the information from the input feature maps due to the group mechanism; compared with grouped convolution, MSGC benefits from the learnability, the core of channel pruning, for constructing its group topology, leading to better channel division. The middle spectrum area is unfolded along four dimensions: group-wise, layer-wise, sample-wise, and attention-wise, making it possible to reveal more powerful and interpretable structures. As a result, the proposed module acts as a booster that can reduce the computational cost of the host backbones for general image recognition with even improved predictive accuracy. For example, in the experiments on ImageNet dataset for image classification, MSGC can reduce the multiply-accumulates (MACs) of ResNet-18 and ResNet-50 by half but still increase the Top-1 accuracy by more than 1%. With 35% reduction of MACs, MSGC can also increase the Top-1 accuracy of the MobileNetV2 backbone. Results on MS COCO dataset for object detection show similar observations. Our code and trained models are available at https://github.com/hellozhuo/msgc.

Abstract: Large-scale pre-training has brought unimodal fields such as computer vision and natural language processing to a new era. Following this trend, the size of multi-modal learning models constantly increases, leading to an urgent need to reduce the massive computational cost of finetuning these models for downstream tasks. In this paper, we propose an efficient and flexible multimodal fusion method, namely PMF, tailored for fusing unimodally pre-trained transformers. Specifically, we first present a modular multimodal fusion framework that exhibits high flexibility and facilitates mutual interactions among different modalities. In addition, we disentangle vanilla prompts into three types in order to learn different optimizing objectives for multimodal learning. It is also worth noting that we propose to add prompt vectors only on the deep layers of the unimodal transformers, thus significantly reducing the training memory usage. Experiment results show that our proposed method achieves comparable performance to several other multimodal finetuning methods with less than 3% trainable parameters and up to 66% saving of training memory usage.

Abstract: Few-shot learning (FSL) via customization of a deep learning network with limited data has emerged as a promising technique to achieve personalized user experiences on edge devices. However, existing FSL methods primarily assume independent and identically distributed (IID) data and utilize either computational backpropagation updates for each task or a common model with task-specific prototypes. Unfortunately, the former solution is infeasible for edge devices that lack on-device backpropagation capabilities, while the latter often struggles with limited generalization ability, especially for out-of-distribution (OOD) data. This paper proposes a lightweight, plug-and-play FSL module called Task-aware Normalization (TANO) that enables efficient and task-aware adaptation of a deep neural network without backpropagation. TANO covers the properties of multiple user groups by coordinating the updates of several groups of the normalization statistics during meta-training and automatically identifies the appropriate normalization group for a downstream few-shot task. Consequently, TANO provides stable but task-specific estimations of the normalization statistics to close the distribution gaps and achieve efficient model adaptation. Results on both intra-domain and out-of-domain generalization experiments demonstrate that TANO outperforms recent methods in terms of accuracy, inference speed, and model size. Moreover, TANO achieves promising results on widely-used FSL benchmarks and data from real applications.

Abstract: Light fields are a type of image data that capture both spatial and angular scene information by recording light rays emitted by a scene from different orientations. In this context, spatial information is defined as features that remain static regardless of perspectives, while angular information refers to features that vary between viewpoints. We propose a novel neural network that, by design, can separate angular and spatial information of a light field. The network represents spatial information using spatial kernels shared among all Sub-Aperture Images (SAIs), and angular information using sets of angular kernels for each SAI. To further improve the representation capability of the network without increasing parameter number, we also introduce angular kernel allocation and kernel tensor decomposition mechanisms. Extensive experiments demonstrate the benefits of information separation: when applied to the compression task, our network outperforms other state-of-the-art methods by a large margin. And angular information can be easily transferred to other scenes for rendering dense views, showing the successful separation and the potential use case for the view synthesis task. We plan to release the code upon acceptance of the paper to encourage further research on this topic.

Abstract: Monocular depth estimation is fundamental for 3D scene understanding and downstream applications. However, even under the supervised setup, it is still challenging and ill-posed due to the lack of full geometric constraints. Although a scene can consist of millions of pixels, there are fewer high-level patterns. We propose iDisc to learn those patterns with internal discretized representations. The method implicitly partitions the scene into a set of high-level patterns. In particular, our new module, Internal Discretization (ID), implements a continuous-discrete-continuous bottleneck to learn those concepts without supervision. In contrast to state-of-the-art methods, the proposed model does not enforce any explicit constraints or priors on the depth output. The whole network with the ID module can be trained end-to-end, thanks to the bottleneck module based on attention. Our method sets the new state of the art with significant improvements on NYU-Depth v2 and KITTI, outperforming all published methods on the official KITTI benchmark. iDisc can also achieve state-of-the-art results on surface normal estimation. Further, we explore the model generalization capability via zero-shot testing. We observe the compelling need to promote diversification in the outdoor scenario. Hence, we introduce splits of two autonomous driving datasets, DDAD and Argoverse. Code is available at http://vis.xyz/pub/idisc .

Abstract: The dominant majority of 3D models that appear in gaming, VR/AR, and those we use to train geometric deep learning algorithms are incomplete, since they are modeled as surface meshes and missing their interior structures. We present a learning framework to recover the shape interiors (RoSI) of existing 3D models with only their exteriors from multi-view and multi-articulation images. Given a set of RGB images that capture a target 3D object in different articulated poses, possibly from only few views, our method infers the interior planes that are observable in the input images. Our neural architecture is trained in a category-agnostic manner and it consists of a motion-aware multi-view analysis phase including pose, depth, and motion estimations, followed by interior plane detection in images and 3D space, and finally multi-view plane fusion. In addition, our method also predicts part articulations and is able to realize and even extrapolate the captured motions on the target 3D object. We evaluate our method by quantitative and qualitative comparisons to baselines and alternative solutions, as well as testing on untrained object categories and real image inputs to assess its generalization capabilities.

Abstract: The structural re-parameterization (SRP) technique is a novel deep learning technique that achieves interconversion between different network architectures through equivalent parameter transformations. This technique enables the mitigation of the extra costs for performance improvement during training, such as parameter size and inference time, through these transformations during inference, and therefore SRP has great potential for industrial and practical applications. The existing SRP methods have successfully considered many commonly used architectures, such as normalizations, pooling methods, multi-branch convolution. However, the widely used self-attention modules cannot be directly implemented by SRP due to these modules usually act on the backbone network in a multiplicative manner and the modules' output is input-dependent during inference, which limits the application scenarios of SRP. In this paper, we conduct extensive experiments from a statistical perspective and discover an interesting phenomenon Stripe Observation, which reveals that channel attention values quickly approach some constant vectors during training. This observation inspires us to propose a simple-yet-effective attention-alike structural re-parameterization (ASR) that allows us to achieve SRP for a given network while enjoying the effectiveness of the self-attention mechanism. Extensive experiments conducted on several standard benchmarks demonstrate the effectiveness of ASR in generally improving the performance of existing backbone networks, self-attention modules, and SRP methods without any elaborated model crafting. We also analyze the limitations and provide experimental or theoretical evidence for the strong robustness of the proposed ASR.

Abstract: Transformer is beneficial for image denoising tasks since it can model long-range dependencies to overcome the limitations presented by inductive convolutional biases. However, directly applying the transformer structure to remove noise is challenging because its complexity grows quadratically with the spatial resolution. In this paper, we propose an efficient Dual-branch Deformable Transformer (DDT) denoising network which captures both local and global interactions in parallel. We divide features with a fixed patch size and a fixed number of patches in local and global branches, respectively. In addition, we apply deformable attention operation in both branches, which helps the network focus on more important regions and further reduces computational complexity. We conduct extensive experiments on real-world and synthetic denoising tasks, and the proposed DDT achieves state-of-the-art performance with significantly fewer computational costs.

Abstract: Recently, event cameras have shown large applicability in several computer vision fields especially concerning tasks that require high temporal resolution. In this work, we investigate the usage of such kind of data for emotion recognition by presenting NEFER, a dataset for Neuromorphic Event-based Facial Expression Recognition. NEFER is composed of paired RGB and event videos representing human faces labeled with the respective emotions and also annotated with face bounding boxes and facial landmarks. We detail the data acquisition process as well as providing a baseline method for RGB and event data. The collected data captures subtle micro-expressions, which are hard to spot with RGB data, yet emerge in the event domain. We report a double recognition accuracy for the event-based approach, proving the effectiveness of a neuromorphic approach for analyzing fast and hardly detectable expressions and the emotions they conceal.

Abstract: We propose the gradient-weighted Object Detector Activation Maps (ODAM), a visualized explanation technique for interpreting the predictions of object detectors. Utilizing the gradients of detector targets flowing into the intermediate feature maps, ODAM produces heat maps that show the influence of regions on the detector's decision for each predicted attribute. Compared to previous works classification activation maps (CAM), ODAM generates instance-specific explanations rather than class-specific ones. We show that ODAM is applicable to both one-stage detectors and two-stage detectors with different types of detector backbones and heads, and produces higher-quality visual explanations than the state-of-the-art both effectively and efficiently. We next propose a training scheme, Odam-Train, to improve the explanation ability on object discrimination of the detector through encouraging consistency between explanations for detections on the same object, and distinct explanations for detections on different objects. Based on the heat maps produced by ODAM with Odam-Train, we propose Odam-NMS, which considers the information of the model's explanation for each prediction to distinguish the duplicate detected objects. We present a detailed analysis of the visualized explanations of detectors and carry out extensive experiments to validate the effectiveness of the proposed ODAM.

Abstract: Learning the hash representation of multi-view heterogeneous data is an important task in multimedia retrieval. However, existing methods fail to effectively fuse the multi-view features and utilize the metric information provided by the dissimilar samples, leading to limited retrieval precision. Current methods utilize weighted sum or concatenation to fuse the multi-view features. We argue that these fusion methods cannot capture the interaction among different views. Furthermore, these methods ignored the information provided by the dissimilar samples. We propose a novel deep metric multi-view hashing (DMMVH) method to address the mentioned problems. Extensive empirical evidence is presented to show that gate-based fusion is better than typical methods. We introduce deep metric learning to the multi-view hashing problems, which can utilize metric information of dissimilar samples. On the MIR-Flickr25K, MS COCO, and NUS-WIDE, our method outperforms the current state-of-the-art methods by a large margin (up to 15.28 mean Average Precision (mAP) improvement).

Abstract: Driver Monitoring Systems (DMSs) are crucial for safe hand-over actions in Level-2+ self-driving vehicles. State-of-the-art DMSs leverage multiple sensors mounted at different locations to monitor the driver and the vehicle's interior scene and employ decision-level fusion to integrate these heterogenous data. However, this fusion method may not fully utilize the complementarity of different data sources and may overlook their relative importance. To address these limitations, we propose a novel multiview multimodal driver monitoring system based on feature-level fusion through multi-head self-attention (MHSA). We demonstrate its effectiveness by comparing it against four alternative fusion strategies (Sum, Conv, SE, and AFF). We also present a novel GPU-friendly supervised contrastive learning framework SuMoCo to learn better representations. Furthermore, We fine-grained the test split of the DAD dataset to enable the multi-class recognition of drivers' activities. Experiments on this enhanced database demonstrate that 1) the proposed MHSA-based fusion method (AUC-ROC: 97.0\%) outperforms all baselines and previous approaches, and 2) training MHSA with patch masking can improve its robustness against modality/view collapses. The code and annotations are publicly available.

Abstract: We introduce Flatlandia, a novel problem for visual localization of an image from object detections composed of two specific tasks: i) Coarse Map Localization: localizing a single image observing a set of objects in respect to a 2D map of object landmarks; ii) Fine-grained 3DoF Localization: estimating latitude, longitude, and orientation of the image within a 2D map. Solutions for these new tasks exploit the wide availability of open urban maps annotated with GPS locations of common objects (\eg via surveying or crowd-sourced). Such maps are also more storage-friendly than standard large-scale 3D models often used in visual localization while additionally being privacy-preserving. As existing datasets are unsuited for the proposed problem, we provide the Flatlandia dataset, designed for 3DoF visual localization in multiple urban settings and based on crowd-sourced data from five European cities. We use the Flatlandia dataset to validate the complexity of the proposed tasks.

Abstract: It is well known that a band-limited signal can be reconstructed from its uniformly spaced samples if the sampling rate is sufficiently high. More recently, it has been proved that one can reconstruct a 1D band-limited signal even if the exact sample locations are unknown, but given just the distribution of the sample locations and their ordering in 1D. In this work, we extend the analytical bounds on the reconstruction error in such scenarios for quasi-bandlimited signals. We also prove that the method for such a reconstruction is resilient to a certain proportion of errors in the specification of the sample location ordering. We then express the problem of tomographic reconstruction of 2D images from 1D Radon projections under unknown angles with known angle distribution, as a special case for reconstruction of quasi-bandlimited signals from samples at unknown locations with known distribution. Building upon our theoretical background, we present asymptotic bounds for 2D quasi-bandlimited image reconstruction from 1D Radon projections in the unknown angles setting, which commonly occurs in cryo-electron microscopy (cryo-EM). To the best of our knowledge, this is the first piece of work to perform such an analysis for 2D cryo-EM, even though the associated reconstruction algorithms have been known for a long time.

Abstract: This paper explores a hierarchical prompting mechanism for the hierarchical image classification (HIC) task. Different from prior HIC methods, our hierarchical prompting is the first to explicitly inject ancestor-class information as a tokenized hint that benefits the descendant-class discrimination. We think it well imitates human visual recognition, i.e., humans may use the ancestor class as a prompt to draw focus on the subtle differences among descendant classes. We model this prompting mechanism into a Transformer with Hierarchical Prompting (TransHP). TransHP consists of three steps: 1) learning a set of prompt tokens to represent the coarse (ancestor) classes, 2) on-the-fly predicting the coarse class of the input image at an intermediate block, and 3) injecting the prompt token of the predicted coarse class into the intermediate feature. Though the parameters of TransHP maintain the same for all input images, the injected coarse-class prompt conditions (modifies) the subsequent feature extraction and encourages a dynamic focus on relatively subtle differences among the descendant classes. Extensive experiments show that TransHP improves image classification on accuracy (e.g., improving ViT-B/16 by +2.83% ImageNet classification accuracy), training data efficiency (e.g., +12.69% improvement under 10% ImageNet training data), and model explainability. Moreover, TransHP also performs favorably against prior HIC methods, showing that TransHP well exploits the hierarchical information.

Abstract: The lack of interpretability of the Vision Transformer may hinder its use in critical real-world applications despite its effectiveness. To overcome this issue, we propose a post-hoc interpretability method called VISION DIFFMASK, which uses the activations of the model's hidden layers to predict the relevant parts of the input that contribute to its final predictions. Our approach uses a gating mechanism to identify the minimal subset of the original input that preserves the predicted distribution over classes. We demonstrate the faithfulness of our method, by introducing a faithfulness task, and comparing it to other state-of-the-art attribution methods on CIFAR-10 and ImageNet-1K, achieving compelling results. To aid reproducibility and further extension of our work, we open source our implementation: https://github.com/AngelosNal/Vision-DiffMask

Abstract: In this paper, we propose an ultrafast automated model compression framework called SeerNet for flexible network deployment. Conventional non-differen-tiable methods discretely search the desirable compression policy based on the accuracy from exhaustively trained lightweight models, and existing differentiable methods optimize an extremely large supernet to obtain the required compressed model for deployment. They both cause heavy computational cost due to the complex compression policy search and evaluation process. On the contrary, we obtain the optimal efficient networks by directly optimizing the compression policy with an accurate performance predictor, where the ultrafast automated model compression for various computational cost constraint is achieved without complex compression policy search and evaluation. Specifically, we first train the performance predictor based on the accuracy from uncertain compression policies actively selected by efficient evolutionary search, so that informative supervision is provided to learn the accurate performance predictor with acceptable cost. Then we leverage the gradient that maximizes the predicted performance under the barrier complexity constraint for ultrafast acquisition of the desirable compression policy, where adaptive update stepsizes with momentum are employed to enhance optimality of the acquired pruning and quantization strategy. Compared with the state-of-the-art automated model compression methods, experimental results on image classification and object detection show that our method achieves competitive accuracy-complexity trade-offs with significant reduction of the search cost.

Abstract: More information leads to better decisions and predictions, right? Confirming this hypothesis, several studies concluded that the simultaneous use of optical and thermal images leads to better predictions in crowd counting. However, the way multimodal models extract enriched features from both modalities is not yet fully understood. Since the use of multimodal data usually increases the complexity, inference time, and memory requirements of the models, it is relevant to examine the differences and advantages of multimodal compared to monomodal models. In this work, all available multimodal datasets for crowd counting are used to investigate the differences between monomodal and multimodal models. To do so, we designed a monomodal architecture that considers the current state of research on monomodal crowd counting. In addition, several multimodal architectures have been developed using different multimodal learning strategies. The key components of the monomodal architecture are also used in the multimodal architectures to be able to answer whether multimodal models perform better in crowd counting in general. Surprisingly, no general answer to this question can be derived from the existing datasets. We found that the existing datasets hold a bias toward thermal images. This was determined by analyzing the relationship between the brightness of optical images and crowd count as well as examining the annotations made for each dataset. Since answering this question is important for future real-world applications of crowd counting, this paper establishes criteria for a potential dataset suitable for answering whether multimodal models perform better in crowd counting in general.

Abstract: In this paper, we propose a novel fully unsupervised framework that learns action representations suitable for the action segmentation task from the single input video itself, without requiring any training data. Our method is a deep metric learning approach rooted in a shallow network with a triplet loss operating on similarity distributions and a novel triplet selection strategy that effectively models temporal and semantic priors to discover actions in the new representational space. Under these circumstances, we successfully recover temporal boundaries in the learned action representations with higher quality compared with existing unsupervised approaches. The proposed method is evaluated on two widely used benchmark datasets for the action segmentation task and it achieves competitive performance by applying a generic clustering algorithm on the learned representations.

Abstract: Detecting fake images is becoming a major goal of computer vision. This need is becoming more and more pressing with the continuous improvement of synthesis methods based on Generative Adversarial Networks (GAN), and even more with the appearance of powerful methods based on Diffusion Models (DM). Towards this end, it is important to gain insight into which image features better discriminate fake images from real ones. In this paper we report on our systematic study of a large number of image generators of different families, aimed at discovering the most forensically relevant characteristics of real and generated images. Our experiments provide a number of interesting observations and shed light on some intriguing properties of synthetic images: (1) not only the GAN models but also the DM and VQ-GAN (Vector Quantized Generative Adversarial Networks) models give rise to visible artifacts in the Fourier domain and exhibit anomalous regular patterns in the autocorrelation; (2) when the dataset used to train the model lacks sufficient variety, its biases can be transferred to the generated images; (3) synthetic and real images exhibit significant differences in the mid-high frequency signal content, observable in their radial and angular spectral power distributions.

Abstract: Predicting high-fidelity future human poses, from a historically observed sequence, is decisive for intelligent robots to interact with humans. Deep end-to-end learning approaches, which typically train a generic pre-trained model on external datasets and then directly apply it to all test samples, emerge as the dominant solution to solve this issue. Despite encouraging progress, they remain non-optimal, as the unique properties (e.g., motion style, rhythm) of a specific sequence cannot be adapted. More generally, at test-time, once encountering unseen motion categories (out-of-distribution), the predicted poses tend to be unreliable. Motivated by this observation, we propose a novel test-time adaptation framework that leverages two self-supervised auxiliary tasks to help the primary forecasting network adapt to the test sequence. In the testing phase, our model can adjust the model parameters by several gradient updates to improve the generation quality. However, due to catastrophic forgetting, both auxiliary tasks typically tend to the low ability to automatically present the desired positive incentives for the final prediction performance. For this reason, we also propose a meta-auxiliary learning scheme for better adaptation. In terms of general setup, our approach obtains higher accuracy, and under two new experimental designs for out-of-distribution data (unseen subjects and categories), achieves significant improvements.

Abstract: Most model-free visual object tracking methods formulate the tracking task as object location estimation given by a 2D segmentation or a bounding box in each video frame. We argue that this representation is limited and instead propose to guide and improve 2D tracking with an explicit object representation, namely the textured 3D shape and 6DoF pose in each video frame. Our representation tackles a complex long-term dense correspondence problem between all 3D points on the object for all video frames, including frames where some points are invisible. To achieve that, the estimation is driven by re-rendering the input video frames as well as possible through differentiable rendering, which has not been used for tracking before. The proposed optimization minimizes a novel loss function to estimate the best 3D shape, texture, and 6DoF pose. We improve the state-of-the-art in 2D segmentation tracking on three different datasets with mostly rigid objects.

Abstract: Certified defense methods against adversarial perturbations have been recently investigated in the black-box setting with a zeroth-order (ZO) perspective. However, these methods suffer from high model variance with low performance on high-dimensional datasets due to the ineffective design of the denoiser and are limited in their utilization of ZO techniques. To this end, we propose a certified ZO preprocessing technique for removing adversarial perturbations from the attacked image in the black-box setting using only model queries. We propose a robust UNet denoiser (RDUNet) that ensures the robustness of black-box models trained on high-dimensional datasets. We propose a novel black-box denoised smoothing (DS) defense mechanism, ZO-RUDS, by prepending our RDUNet to the black-box model, ensuring black-box defense. We further propose ZO-AE-RUDS in which RDUNet followed by autoencoder (AE) is prepended to the black-box model. We perform extensive experiments on four classification datasets, CIFAR-10, CIFAR-10, Tiny Imagenet, STL-10, and the MNIST dataset for image reconstruction tasks. Our proposed defense methods ZO-RUDS and ZO-AE-RUDS beat SOTA with a huge margin of $35\%$ and $9\%$, for low dimensional (CIFAR-10) and with a margin of $20.61\%$ and $23.51\%$ for high-dimensional (STL-10) datasets, respectively.

Abstract: We propose a novel method for Zero-Shot Anomaly Localization that leverages a bidirectional mapping derived from the 1-dimensional Wasserstein Distance. The proposed approach allows pinpointing the anomalous regions in a texture with increased precision by aggregating the contribution of a pixel to the errors of all nearby patches. We validate our solution on several datasets and obtain more than a 40% reduction in error over the previous state of the art on the MVTec AD dataset in a zero-shot setting.

Abstract: Video quality assessment (VQA) aims to simulate the human perception of video quality, which is influenced by factors ranging from low-level color and texture details to high-level semantic content. To effectively model these complicated quality-related factors, in this paper, we decompose video into three levels (\ie, patch level, frame level, and clip level), and propose a novel Zoom-VQA architecture to perceive spatio-temporal features at different levels. It integrates three components: patch attention module, frame pyramid alignment, and clip ensemble strategy, respectively for capturing region-of-interest in the spatial dimension, multi-level information at different feature levels, and distortions distributed over the temporal dimension. Owing to the comprehensive design, Zoom-VQA obtains state-of-the-art results on four VQA benchmarks and achieves 2nd place in the NTIRE 2023 VQA challenge. Notably, Zoom-VQA has outperformed the previous best results on two subsets of LSVQ, achieving 0.8860 (+1.0%) and 0.7985 (+1.9%) of SRCC on the respective subsets. Adequate ablation studies further verify the effectiveness of each component. Codes and models are released in https://github.com/k-zha14/Zoom-VQA.

Abstract: Vision transformers have been applied successfully for image recognition tasks. There have been either multi-headed self-attention based (ViT \cite{dosovitskiy2020image}, DeIT, \cite{touvron2021training}) similar to the original work in textual models or more recently based on spectral layers (Fnet\cite{lee2021fnet}, GFNet\cite{rao2021global}, AFNO\cite{guibas2021efficient}). We hypothesize that both spectral and multi-headed attention plays a major role. We investigate this hypothesis through this work and observe that indeed combining spectral and multi-headed attention layers provides a better transformer architecture. We thus propose the novel Spectformer architecture for transformers that combines spectral and multi-headed attention layers. We believe that the resulting representation allows the transformer to capture the feature representation appropriately and it yields improved performance over other transformer representations. For instance, it improves the top-1 accuracy by 2\% on ImageNet compared to both GFNet-H and LiT. SpectFormer-S reaches 84.25\% top-1 accuracy on ImageNet-1K (state of the art for small version). Further, Spectformer-L achieves 85.7\% that is the state of the art for the comparable base version of the transformers. We further ensure that we obtain reasonable results in other scenarios such as transfer learning on standard datasets such as CIFAR-10, CIFAR-100, Oxford-IIIT-flower, and Standford Car datasets. We then investigate its use in downstream tasks such of object detection and instance segmentation on the MS-COCO dataset and observe that Spectformer shows consistent performance that is comparable to the best backbones and can be further optimized and improved. Hence, we believe that combined spectral and attention layers are what are needed for vision transformers.

Abstract: Document-based Visual Question Answering examines the document understanding of document images in conditions of natural language questions. We proposed a new document-based VQA dataset, PDF-VQA, to comprehensively examine the document understanding from various aspects, including document element recognition, document layout structural understanding as well as contextual understanding and key information extraction. Our PDF-VQA dataset extends the current scale of document understanding that limits on the single document page to the new scale that asks questions over the full document of multiple pages. We also propose a new graph-based VQA model that explicitly integrates the spatial and hierarchically structural relationships between different document elements to boost the document structural understanding. The performances are compared with several baselines over different question types and tasks\footnote{The full dataset will be released after paper acceptance.

Abstract: With the development of high-definition display devices, the practical scenario of Super-Resolution (SR) usually needs to super-resolve large input like 2K to higher resolution (4K/8K). To reduce the computational and memory cost, current methods first split the large input into local patches and then merge the SR patches into the output. These methods adaptively allocate a subnet for each patch. Quantization is a very important technique for network acceleration and has been used to design the subnets. Current methods train an MLP bit selector to determine the propoer bit for each layer. However, they uniformly sample subnets for training, making simple subnets overfitted and complicated subnets underfitted. Therefore, the trained bit selector fails to determine the optimal bit. Apart from this, the introduced bit selector brings additional cost to each layer of the SR network. In this paper, we propose a novel method named Content-Aware Bit Mapping (CABM), which can remove the bit selector without any performance loss. CABM also learns a bit selector for each layer during training. After training, we analyze the relation between the edge information of an input patch and the bit of each layer. We observe that the edge information can be an effective metric for the selected bit. Therefore, we design a strategy to build an Edge-to-Bit lookup table that maps the edge score of a patch to the bit of each layer during inference. The bit configuration of SR network can be determined by the lookup tables of all layers. Our strategy can find better bit configuration, resulting in more efficient mixed precision networks. We conduct detailed experiments to demonstrate the generalization ability of our method. The code will be released.

Abstract: Self-supervised learning (SSL) has made remarkable progress in visual representation learning. Some studies combine SSL with knowledge distillation (SSL-KD) to boost the representation learning performance of small models. In this study, we propose a Multi-mode Online Knowledge Distillation method (MOKD) to boost self-supervised visual representation learning. Different from existing SSL-KD methods that transfer knowledge from a static pre-trained teacher to a student, in MOKD, two different models learn collaboratively in a self-supervised manner. Specifically, MOKD consists of two distillation modes: self-distillation and cross-distillation modes. Among them, self-distillation performs self-supervised learning for each model independently, while cross-distillation realizes knowledge interaction between different models. In cross-distillation, a cross-attention feature search strategy is proposed to enhance the semantic feature alignment between different models. As a result, the two models can absorb knowledge from each other to boost their representation learning performance. Extensive experimental results on different backbones and datasets demonstrate that two heterogeneous models can benefit from MOKD and outperform their independently trained baseline. In addition, MOKD also outperforms existing SSL-KD methods for both the student and teacher models.

Abstract: Super-Resolution (SR) has gained increasing research attention over the past few years. With the development of Deep Neural Networks (DNNs), many super-resolution methods based on DNNs have been proposed. Although most of these methods are aimed at ordinary frames, there are few works on super-resolution of omnidirectional frames. In these works, omnidirectional frames are projected from the 3D sphere to a 2D plane by Equi-Rectangular Projection (ERP). Although ERP has been widely used for projection, it has severe projection distortion near poles. Current DNN-based SR methods use 2D convolution modules, which is more suitable for the regular grid. In this paper, we find that different projection methods have great impact on the performance of DNNs. To study this problem, a comprehensive comparison of projections in omnidirectional super-resolution is conducted. We compare the SR results of different projection methods. Experimental results show that Equi-Angular cube map projection (EAC), which has minimal distortion, achieves the best result in terms of WS-PSNR compared with other projections. Code and data will be released.

Abstract: 3D scanning as a technique to digitize objects in reality and create their 3D models, is used in many fields and areas. Though the quality of 3D scans depends on the technical characteristics of the 3D scanner, the common drawback is the smoothing of fine details, or the edges of an object. We introduce SepicNet, a novel deep network for the detection and parametrization of sharp edges in 3D shapes as primitive curves. To make the network end-to-end trainable, we formulate the curve fitting in a differentiable manner. We develop an adaptive point cloud sampling technique that captures the sharp features better than uniform sampling. The experiments were conducted on a newly introduced large-scale dataset of 50k 3D scans, where the sharp edge annotations were extracted from their parametric CAD models, and demonstrate significant improvement over state-of-the-art methods.

Abstract: This paper proposes a novel method for human hands tracking using data from an event camera. The event camera detects changes in brightness, measuring motion, with low latency, no motion blur, low power consumption and high dynamic range. Captured frames are analysed using lightweight algorithms reporting 3D hand position data. The chosen pick-and-place scenario serves as an example input for collaborative human-robot interactions and in obstacle avoidance for human-robot safety applications. Events data are pre-processed into intensity frames. The regions of interest (ROI) are defined through object edge event activity, reducing noise. ROI features are extracted for use in-depth perception. Event-based tracking of human hand demonstrated feasible, in real time and at a low computational cost. The proposed ROI-finding method reduces noise from intensity images, achieving up to 89% of data reduction in relation to the original, while preserving the features. The depth estimation error in relation to ground truth (measured with wearables), measured using dynamic time warping and using a single event camera, is from 15 to 30 millimetres, depending on the plane it is measured. Tracking of human hands in 3D space using a single event camera data and lightweight algorithms to define ROI features (hands tracking in space).

Abstract: How to estimate the uncertainty of a given model is a crucial problem. Current calibration techniques treat different classes equally and thus implicitly assume that the distribution of training data is balanced, but ignore the fact that real-world data often follows a long-tailed distribution. In this paper, we explore the problem of calibrating the model trained from a long-tailed distribution. Due to the difference between the imbalanced training distribution and balanced test distribution, existing calibration methods such as temperature scaling can not generalize well to this problem. Specific calibration methods for domain adaptation are also not applicable because they rely on unlabeled target domain instances which are not available. Models trained from a long-tailed distribution tend to be more overconfident to head classes. To this end, we propose a novel knowledge-transferring-based calibration method by estimating the importance weights for samples of tail classes to realize long-tailed calibration. Our method models the distribution of each class as a Gaussian distribution and views the source statistics of head classes as a prior to calibrate the target distributions of tail classes. We adaptively transfer knowledge from head classes to get the target probability density of tail classes. The importance weight is estimated by the ratio of the target probability density over the source probability density. Extensive experiments on CIFAR-10-LT, MNIST-LT, CIFAR-100-LT, and ImageNet-LT datasets demonstrate the effectiveness of our method.

Abstract: Existing implicit neural representation (INR) methods do not fully exploit spatiotemporal redundancies in videos. Index-based INRs ignore the content-specific spatial features and hybrid INRs ignore the contextual dependency on adjacent frames, leading to poor modeling capability for scenes with large motion or dynamics. We analyze this limitation from the perspective of function fitting and reveal the importance of frame difference. To use explicit motion information, we propose Difference Neural Representation for Videos (DNeRV), which consists of two streams for content and frame difference. We also introduce a collaborative content unit for effective feature fusion. We test DNeRV for video compression, inpainting, and interpolation. DNeRV achieves competitive results against the state-of-the-art neural compression approaches and outperforms existing implicit methods on downstream inpainting and interpolation for $960 \times 1920$ videos.

Abstract: A reliable perception has to be robust against challenging environmental conditions. Therefore, recent efforts focused on the use of radar sensors in addition to camera and lidar sensors for perception applications. However, the sparsity of radar point clouds and the poor data availability remain challenging for current perception methods. To address these challenges, a novel graph neural network is proposed that does not just use the information of the points themselves but also the relationships between the points. The model is designed to consider both point features and point-pair features, embedded in the edges of the graph. Furthermore, a general approach for achieving transformation invariance is proposed which is robust against unseen scenarios and also counteracts the limited data availability. The transformation invariance is achieved by an invariant data representation rather than an invariant model architecture, making it applicable to other methods. The proposed RadarGNN model outperforms all previous methods on the RadarScenes dataset. In addition, the effects of different invariances on the object detection and semantic segmentation quality are investigated. The code is made available as open-source software under https://github.com/TUMFTM/RadarGNN.

Abstract: This paper proposes a novel approach to real-time automatic rim detection, classification, and inspection by combining traditional computer vision and deep learning techniques. At the end of every automotive assembly line, a quality control process is carried out to identify any potential defects in the produced cars. Common yet hazardous defects are related, for example, to incorrectly mounted rims. Routine inspections are mostly conducted by human workers that are negatively affected by factors such as fatigue or distraction. We have designed a new prototype to validate whether all four wheels on a single car match in size and type. Additionally, we present three comprehensive open-source databases, CWD1500, WHEEL22, and RB600, for wheel, rim, and bolt detection, as well as rim classification, which are free-to-use for scientific purposes.

Abstract: We introduce NeRD, a new demosaicking method for generating full-color images from Bayer patterns. Our approach leverages advancements in neural fields to perform demosaicking by representing an image as a coordinate-based neural network with sine activation functions. The inputs to the network are spatial coordinates and a low-resolution Bayer pattern, while the outputs are the corresponding RGB values. An encoder network, which is a blend of ResNet and U-net, enhances the implicit neural representation of the image to improve its quality and ensure spatial consistency through prior learning. Our experimental results demonstrate that NeRD outperforms traditional and state-of-the-art CNN-based methods and significantly closes the gap to transformer-based methods.

Abstract: Age is an important variable to describe the expected brain's anatomy status across the normal aging trajectory. The deviation from that normative aging trajectory may provide some insights into neurological diseases. In neuroimaging, predicted brain age is widely used to analyze different diseases. However, using only the brain age gap information (\ie the difference between the chronological age and the estimated age) can be not enough informative for disease classification problems. In this paper, we propose to extend the notion of global brain age by estimating brain structure ages using structural magnetic resonance imaging. To this end, an ensemble of deep learning models is first used to estimate a 3D aging map (\ie voxel-wise age estimation). Then, a 3D segmentation mask is used to obtain the final brain structure ages. This biomarker can be used in several situations. First, it enables to accurately estimate the brain age for the purpose of anomaly detection at the population level. In this situation, our approach outperforms several state-of-the-art methods. Second, brain structure ages can be used to compute the deviation from the normal aging process of each brain structure. This feature can be used in a multi-disease classification task for an accurate differential diagnosis at the subject level. Finally, the brain structure age deviations of individuals can be visualized, providing some insights about brain abnormality and helping clinicians in real medical contexts.

Abstract: Humans possess the capacity to reason about the future based on a sparse collection of visual cues acquired over time. In order to emulate this ability, we introduce a novel task called Anticipation Captioning, which generates a caption for an unseen oracle image using a sparsely temporally-ordered set of images. To tackle this new task, we propose a model called A-CAP, which incorporates commonsense knowledge into a pre-trained vision-language model, allowing it to anticipate the caption. Through both qualitative and quantitative evaluations on a customized visual storytelling dataset, A-CAP outperforms other image captioning methods and establishes a strong baseline for anticipation captioning. We also address the challenges inherent in this task.

Abstract: This paper investigates the problem of class-incremental object detection for agricultural applications where a model needs to learn new plant species and diseases incrementally without forgetting the previously learned ones. We adapt two public datasets to include new categories over time, simulating a more realistic and dynamic scenario. We then compare three class-incremental learning methods that leverage different forms of knowledge distillation to mitigate catastrophic forgetting. Our experiments show that all three methods suffer from catastrophic forgetting, but the recent Dynamic Y-KD approach, which additionally uses a dynamic architecture that grows new branches to learn new tasks, outperforms ILOD and Faster-ILOD in most scenarios both on new and old classes. These results highlight the challenges and opportunities of continual object detection for agricultural applications. In particular, the large intra-class and small inter-class variability that is typical of plant images exacerbate the difficulty of learning new categories without interfering with previous knowledge. We publicly release our code to encourage future work.

Abstract: Diffusion models have proven to be highly effective in generating high-quality images. However, adapting large pre-trained diffusion models to new domains remains an open challenge, which is critical for real-world applications. This paper proposes DiffFit, a parameter-efficient strategy to fine-tune large pre-trained diffusion models that enable fast adaptation to new domains. DiffFit is embarrassingly simple that only fine-tunes the bias term and newly-added scaling factors in specific layers, yet resulting in significant training speed-up and reduced model storage costs. Compared with full fine-tuning, DiffFit achieves 2$\times$ training speed-up and only needs to store approximately 0.12\% of the total model parameters. Intuitive theoretical analysis has been provided to justify the efficacy of scaling factors on fast adaptation. On 8 downstream datasets, DiffFit achieves superior or competitive performances compared to the full fine-tuning while being more efficient. Remarkably, we show that DiffFit can adapt a pre-trained low-resolution generative model to a high-resolution one by adding minimal cost. Among diffusion-based methods, DiffFit sets a new state-of-the-art FID of 3.02 on ImageNet 512$\times$512 benchmark by fine-tuning only 25 epochs from a public pre-trained ImageNet 256$\times$256 checkpoint while being 30$\times$ more training efficient than the closest competitor.

Abstract: Due to the limitations of inadequate Whole-Slide Image (WSI) samples with weak labels, pseudo-bag-based multiple instance learning (MIL) appears as a vibrant prospect in WSI classification. However, the pseudo-bag dividing scheme, often crucial for classification performance, is still an open topic worth exploring. Therefore, this paper proposes a novel scheme, ProtoDiv, using a bag prototype to guide the division of WSI pseudo-bags. Rather than designing complex network architecture, this scheme takes a plugin-and-play approach to safely augment WSI data for effective training while preserving sample consistency. Furthermore, we specially devise an attention-based prototype that could be optimized dynamically in training to adapt to a classification task. We apply our ProtoDiv scheme on seven baseline models, and then carry out a group of comparison experiments on two public WSI datasets. Experiments confirm our ProtoDiv could usually bring obvious performance improvements to WSI classification.

Abstract: Most state-of-the-art instance segmentation methods rely on large amounts of pixel-precise ground-truth annotations for training, which are expensive to create. Interactive segmentation networks help generate such annotations based on an image and the corresponding user interactions such as clicks. Existing methods for this task can only process a single instance at a time and each user interaction requires a full forward pass through the entire deep network. We introduce a more efficient approach, called DynaMITe, in which we represent user interactions as spatio-temporal queries to a Transformer decoder with a potential to segment multiple object instances in a single iteration. Our architecture also alleviates any need to re-compute image features during refinement, and requires fewer interactions for segmenting multiple instances in a single image when compared to other methods. DynaMITe achieves state-of-the-art results on multiple existing interactive segmentation benchmarks, and also on the new multi-instance benchmark that we propose in this paper.

Abstract: Spatial control is a core capability in controllable image generation. Advancements in layout-guided image generation have shown promising results on in-distribution (ID) datasets with similar spatial configurations. However, it is unclear how these models perform when facing out-of-distribution (OOD) samples with arbitrary, unseen layouts. In this paper, we propose LayoutBench, a diagnostic benchmark for layout-guided image generation that examines four categories of spatial control skills: number, position, size, and shape. We benchmark two recent representative layout-guided image generation methods and observe that the good ID layout control may not generalize well to arbitrary layouts in the wild (e.g., objects at the boundary). Next, we propose IterInpaint, a new baseline that generates foreground and background regions in a step-by-step manner via inpainting, demonstrating stronger generalizability than existing models on OOD layouts in LayoutBench. We perform quantitative and qualitative evaluation and fine-grained analysis on the four LayoutBench skills to pinpoint the weaknesses of existing models. Lastly, we show comprehensive ablation studies on IterInpaint, including training task ratio, crop&paste vs. repaint, and generation order. Project website: https://layoutbench.github.io

Abstract: Few-shot learning (FSL) techniques seek to learn the underlying patterns in data using fewer samples, analogous to how humans learn from limited experience. In this limited-data scenario, the challenges associated with deep neural networks, such as shortcut learning and texture bias behaviors, are further exacerbated. Moreover, the significance of addressing shortcut learning is not yet fully explored in the few-shot setup. To address these issues, we propose LSFSL, which enforces the model to learn more generalizable features utilizing the implicit prior information present in the data. Through comprehensive analyses, we demonstrate that LSFSL-trained models are less vulnerable to alteration in color schemes, statistical correlations, and adversarial perturbations leveraging the global semantics in the data. Our findings highlight the potential of incorporating relevant priors in few-shot approaches to increase robustness and generalization.

Abstract: Diffusion models have recently become the de-facto approach for generative modeling in the 2D domain. However, extending diffusion models to 3D is challenging due to the difficulties in acquiring 3D ground truth data for training. On the other hand, 3D GANs that integrate implicit 3D representations into GANs have shown remarkable 3D-aware generation when trained only on single-view image datasets. However, 3D GANs do not provide straightforward ways to precisely control image synthesis. To address these challenges, We present Control3Diff, a 3D diffusion model that combines the strengths of diffusion models and 3D GANs for versatile, controllable 3D-aware image synthesis for single-view datasets. Control3Diff explicitly models the underlying latent distribution (optionally conditioned on external inputs), thus enabling direct control during the diffusion process. Moreover, our approach is general and applicable to any type of controlling input, allowing us to train it with the same diffusion objective without any auxiliary supervision. We validate the efficacy of Control3Diff on standard image generation benchmarks, including FFHQ, AFHQ, and ShapeNet, using various conditioning inputs such as images, sketches, and text prompts. Please see the project website (\url{https://jiataogu.me/control3diff}) for video comparisons.

Abstract: Burst image processing is becoming increasingly popular in recent years. However, it is a challenging task since individual burst images undergo multiple degradations and often have mutual misalignments resulting in ghosting and zipper artifacts. Existing burst restoration methods usually do not consider the mutual correlation and non-local contextual information among burst frames, which tends to limit these approaches in challenging cases. Another key challenge lies in the robust up-sampling of burst frames. The existing up-sampling methods cannot effectively utilize the advantages of single-stage and progressive up-sampling strategies with conventional and/or recent up-samplers at the same time. To address these challenges, we propose a novel Gated Multi-Resolution Transfer Network (GMTNet) to reconstruct a spatially precise high-quality image from a burst of low-quality raw images. GMTNet consists of three modules optimized for burst processing tasks: Multi-scale Burst Feature Alignment (MBFA) for feature denoising and alignment, Transposed-Attention Feature Merging (TAFM) for multi-frame feature aggregation, and Resolution Transfer Feature Up-sampler (RTFU) to up-scale merged features and construct a high-quality output image. Detailed experimental analysis on five datasets validates our approach and sets a state-of-the-art for burst super-resolution, burst denoising, and low-light burst enhancement.

Abstract: We propose a method to estimate the mechanical parameters of fabrics using a casual capture setup with a depth camera. Our approach enables to create mechanically-correct digital representations of real-world textile materials, which is a fundamental step for many interactive design and engineering applications. As opposed to existing capture methods, which typically require expensive setups, video sequences, or manual intervention, our solution can capture at scale, is agnostic to the optical appearance of the textile, and facilitates fabric arrangement by non-expert operators. To this end, we propose a sim-to-real strategy to train a learning-based framework that can take as input one or multiple images and outputs a full set of mechanical parameters. Thanks to carefully designed data augmentation and transfer learning protocols, our solution generalizes to real images despite being trained only on synthetic data, hence successfully closing the sim-to-real loop.Key in our work is to demonstrate that evaluating the regression accuracy based on the similarity at parameter space leads to an inaccurate distances that do not match the human perception. To overcome this, we propose a novel metric for fabric drape similarity that operates on the image domain instead on the parameter space, allowing us to evaluate our estimation within the context of a similarity rank. We show that out metric correlates with human judgments about the perception of drape similarity, and that our model predictions produce perceptually accurate results compared to the ground truth parameters.

Abstract: Neural Radiance Field training can be accelerated through the use of grid-based representations in NeRF's learned mapping from spatial coordinates to colors and volumetric density. However, these grid-based approaches lack an explicit understanding of scale and therefore often introduce aliasing, usually in the form of jaggies or missing scene content. Anti-aliasing has previously been addressed by mip-NeRF 360, which reasons about sub-volumes along a cone rather than points along a ray, but this approach is not natively compatible with current grid-based techniques. We show how ideas from rendering and signal processing can be used to construct a technique that combines mip-NeRF 360 and grid-based models such as Instant NGP to yield error rates that are 8% - 76% lower than either prior technique, and that trains 22x faster than mip-NeRF 360.

Abstract: Recently, there has been an arms race of pose forecasting methods aimed at solving the spatio-temporal task of predicting a sequence of future 3D poses of a person given a sequence of past observed ones. However, the lack of unified benchmarks and limited uncertainty analysis have hindered progress in the field. To address this, we first develop an open-source library for human pose forecasting, featuring multiple models, datasets, and standardized evaluation metrics, with the aim of promoting research and moving toward a unified and fair evaluation. Second, we devise two types of uncertainty in the problem to increase performance and convey better trust: 1) we propose a method for modeling aleatoric uncertainty by using uncertainty priors to inject knowledge about the behavior of uncertainty. This focuses the capacity of the model in the direction of more meaningful supervision while reducing the number of learned parameters and improving stability; 2) we introduce a novel approach for quantifying the epistemic uncertainty of any model through clustering and measuring the entropy of its assignments. Our experiments demonstrate up to $25\%$ improvements in accuracy and better performance in uncertainty estimation.

Abstract: Understanding verbs is crucial to modelling how people and objects interact with each other and the environment through space and time. Recently, state-of-the-art video-language models based on CLIP have been shown to have limited verb understanding and to rely extensively on nouns, restricting their performance in real-world video applications that require action and temporal understanding. In this work, we improve verb understanding for CLIP-based video-language models by proposing a new Verb-Focused Contrastive (VFC) framework. This consists of two main components: (1) leveraging pretrained large language models (LLMs) to create hard negatives for cross-modal contrastive learning, together with a calibration strategy to balance the occurrence of concepts in positive and negative pairs; and (2) enforcing a fine-grained, verb phrase alignment loss. Our method achieves state-of-the-art results for zero-shot performance on three downstream tasks that focus on verb understanding: video-text matching, video question-answering and video classification. To the best of our knowledge, this is the first work which proposes a method to alleviate the verb understanding problem, and does not simply highlight it.

Abstract: Current transformer-based change detection (CD) approaches either employ a pre-trained model trained on large-scale image classification ImageNet dataset or rely on first pre-training on another CD dataset and then fine-tuning on the target benchmark. This current strategy is driven by the fact that transformers typically require a large amount of training data to learn inductive biases, which is insufficient in standard CD datasets due to their small size. We develop an end-to-end CD approach with transformers that is trained from scratch and yet achieves state-of-the-art performance on four public benchmarks. Instead of using conventional self-attention that struggles to capture inductive biases when trained from scratch, our architecture utilizes a shuffled sparse-attention operation that focuses on selected sparse informative regions to capture the inherent characteristics of the CD data. Moreover, we introduce a change-enhanced feature fusion (CEFF) module to fuse the features from input image pairs by performing a per-channel re-weighting. Our CEFF module aids in enhancing the relevant semantic changes while suppressing the noisy ones. Extensive experiments on four CD datasets reveal the merits of the proposed contributions, achieving gains as high as 14.27\% in intersection-over-union (IoU) score, compared to the best-published results in the literature. Code is available at \url{https://github.com/mustansarfiaz/ScratchFormer}.

Abstract: We address the problem of learning person-specific facial priors from a small number (e.g., 20) of portrait photos of the same person. This enables us to edit this specific person's facial appearance, such as expression and lighting, while preserving their identity and high-frequency facial details. Key to our approach, which we dub DiffusionRig, is a diffusion model conditioned on, or "rigged by," crude 3D face models estimated from single in-the-wild images by an off-the-shelf estimator. On a high level, DiffusionRig learns to map simplistic renderings of 3D face models to realistic photos of a given person. Specifically, DiffusionRig is trained in two stages: It first learns generic facial priors from a large-scale face dataset and then person-specific priors from a small portrait photo collection of the person of interest. By learning the CGI-to-photo mapping with such personalized priors, DiffusionRig can "rig" the lighting, facial expression, head pose, etc. of a portrait photo, conditioned only on coarse 3D models while preserving this person's identity and other high-frequency characteristics. Qualitative and quantitative experiments show that DiffusionRig outperforms existing approaches in both identity preservation and photorealism. Please see the project website: https://diffusionrig.github.io for the supplemental material, video, code, and data.

Abstract: Large-scale Vision-Language Models, such as CLIP, learn powerful image-text representations that have found numerous applications, from zero-shot classification to text-to-image generation. Despite that, their capabilities for solving novel discriminative tasks via prompting fall behind those of large language models, such as GPT-3. Here we explore the idea of visual prompt engineering for solving computer vision tasks beyond classification by editing in image space instead of text. In particular, we discover an emergent ability of CLIP, where, by simply drawing a red circle around an object, we can direct the model's attention to that region, while also maintaining global information. We show the power of this simple approach by achieving state-of-the-art in zero-shot referring expressions comprehension and strong performance in keypoint localization tasks. Finally, we draw attention to some potential ethical concerns of large language-vision models.

Abstract: 3D-aware image synthesis encompasses a variety of tasks, such as scene generation and novel view synthesis from images. Despite numerous task-specific methods, developing a comprehensive model remains challenging. In this paper, we present SSDNeRF, a unified approach that employs an expressive diffusion model to learn a generalizable prior of neural radiance fields (NeRF) from multi-view images of diverse objects. Previous studies have used two-stage approaches that rely on pretrained NeRFs as real data to train diffusion models. In contrast, we propose a new single-stage training paradigm with an end-to-end objective that jointly optimizes a NeRF auto-decoder and a latent diffusion model, enabling simultaneous 3D reconstruction and prior learning, even from sparsely available views. At test time, we can directly sample the diffusion prior for unconditional generation, or combine it with arbitrary observations of unseen objects for NeRF reconstruction. SSDNeRF demonstrates robust results comparable to or better than leading task-specific methods in unconditional generation and single/sparse-view 3D reconstruction.

Abstract: Large-scale models pre-trained on large-scale datasets have profoundly advanced the development of deep learning. However, the state-of-the-art models for medical image segmentation are still small-scale, with their parameters only in the tens of millions. Further scaling them up to higher orders of magnitude is rarely explored. An overarching goal of exploring large-scale models is to train them on large-scale medical segmentation datasets for better transfer capacities. In this work, we design a series of Scalable and Transferable U-Net (STU-Net) models, with parameter sizes ranging from 14 million to 1.4 billion. Notably, the 1.4B STU-Net is the largest medical image segmentation model to date. Our STU-Net is based on nnU-Net framework due to its popularity and impressive performance. We first refine the default convolutional blocks in nnU-Net to make them scalable. Then, we empirically evaluate different scaling combinations of network depth and width, discovering that it is optimal to scale model depth and width together. We train our scalable STU-Net models on a large-scale TotalSegmentator dataset and find that increasing model size brings a stronger performance gain. This observation reveals that a large model is promising in medical image segmentation. Furthermore, we evaluate the transferability of our model on 14 downstream datasets for direct inference and 3 datasets for further fine-tuning, covering various modalities and segmentation targets. We observe good performance of our pre-trained model in both direct inference and fine-tuning. The code and pre-trained models are available at https://github.com/Ziyan-Huang/STU-Net.

Abstract: This paper introduces a novel representation of volumetric videos for real-time view synthesis of dynamic scenes. Recent advances in neural scene representations demonstrate their remarkable capability to model and render complex static scenes, but extending them to represent dynamic scenes is not straightforward due to their slow rendering speed or high storage cost. To solve this problem, our key idea is to represent the radiance field of each frame as a set of shallow MLP networks whose parameters are stored in 2D grids, called MLP maps, and dynamically predicted by a 2D CNN decoder shared by all frames. Representing 3D scenes with shallow MLPs significantly improves the rendering speed, while dynamically predicting MLP parameters with a shared 2D CNN instead of explicitly storing them leads to low storage cost. Experiments show that the proposed approach achieves state-of-the-art rendering quality on the NHR and ZJU-MoCap datasets, while being efficient for real-time rendering with a speed of 41.7 fps for $512 \times 512$ images on an RTX 3090 GPU. The code is available at https://zju3dv.github.io/mlp_maps/.

Abstract: Despite the growing demand for interactive AI systems, there have been few comprehensive studies on human-AI interaction in visual understanding e.g. segmentation. Inspired by the development of prompt-based universal interfaces for LLMs, this paper presents SEEM, a promptable, interactive model for Segmenting Everything Everywhere all at once in an image. SEEM has four desiderata: i) Versatility: by introducing a versatile prompting engine for different types of prompts, including points, boxes, scribbles, masks, texts, and referred regions of another image; ii) Compositionality: by learning a joint visual-semantic space for visual and textual prompts to compose queries on the fly for inference as shown in Fig 1; iii)Interactivity: by incorporating learnable memory prompts to retain dialog history information via mask-guided cross-attention; and iv) Semantic-awareness: by using a text encoder to encode text queries and mask labels for open-vocabulary segmentation.

Abstract: The recent advances in camera-based bird's eye view (BEV) representation exhibit great potential for in-vehicle 3D perception. Despite the substantial progress achieved on standard benchmarks, the robustness of BEV algorithms has not been thoroughly examined, which is critical for safe operations. To bridge this gap, we introduce RoboBEV, a comprehensive benchmark suite that encompasses eight distinct corruptions, including Bright, Dark, Fog, Snow, Motion Blur, Color Quant, Camera Crash, and Frame Lost. Based on it, we undertake extensive evaluations across a wide range of BEV-based models to understand their resilience and reliability. Our findings indicate a strong correlation between absolute performance on in-distribution and out-of-distribution datasets. Nonetheless, there are considerable variations in relative performance across different approaches. Our experiments further demonstrate that pre-training and depth-free BEV transformation has the potential to enhance out-of-distribution robustness. Additionally, utilizing long and rich temporal information largely helps with robustness. Our findings provide valuable insights for designing future BEV models that can achieve both accuracy and robustness in real-world deployments.

Abstract: Plain text has become a prevalent interface for text-to-image synthesis. However, its limited customization options hinder users from accurately describing desired outputs. For example, plain text makes it hard to specify continuous quantities, such as the precise RGB color value or importance of each word. Furthermore, creating detailed text prompts for complex scenes is tedious for humans to write and challenging for text encoders to interpret. To address these challenges, we propose using a rich-text editor supporting formats such as font style, size, color, and footnote. We extract each word's attributes from rich text to enable local style control, explicit token reweighting, precise color rendering, and detailed region synthesis. We achieve these capabilities through a region-based diffusion process. We first obtain each word's region based on cross-attention maps of a vanilla diffusion process using plain text. For each region, we enforce its text attributes by creating region-specific detailed prompts and applying region-specific guidance. We present various examples of image generation from rich text and demonstrate that our method outperforms strong baselines with quantitative evaluations.