Computer Vision and Pattern Recognition (cs.CV)

Abstract: Emerging foundation models in machine learning are models trained on vast amounts of data that have been shown to generalize well to new tasks. Often these models can be prompted with multi-modal inputs that range from natural language descriptions over images to point clouds. In this paper, we propose topological data analysis (TDA) guided prompt optimization for the Segment Anything Model (SAM) and show preliminary results in the biological image segmentation domain. Our approach replaces the standard grid search approach that is used in the original implementation and finds point locations based on their topological significance. Our results show that the TDA optimized point cloud is much better suited for finding small objects and massively reduces computational complexity despite the extra step in scenarios which require many segmentations.

Abstract: This technical report describes our QuAVF@NTU-NVIDIA submission to the Ego4D Talking to Me (TTM) Challenge 2023. Based on the observation from the TTM task and the provided dataset, we propose to use two separate models to process the input videos and audio. By doing so, we can utilize all the labeled training data, including those without bounding box labels. Furthermore, we leverage the face quality score from a facial landmark prediction model for filtering noisy face input data. The face quality score is also employed in our proposed quality-aware fusion for integrating the results from two branches. With the simple architecture design, our model achieves 67.4% mean average precision (mAP) on the test set, which ranks first on the leaderboard and outperforms the baseline method by a large margin. Code is available at: https://github.com/hsi-che-lin/Ego4D-QuAVF-TTM-CVPR23

Abstract: Detecting the salient objects in a remote sensing image has wide applications for the interdisciplinary research. Many existing deep learning methods have been proposed for Salient Object Detection (SOD) in remote sensing images and get remarkable results. However, the recent adversarial attack examples, generated by changing a few pixel values on the original remote sensing image, could result in a collapse for the well-trained deep learning based SOD model. Different with existing methods adding perturbation to original images, we propose to jointly tune adversarial exposure and additive perturbation for attack and constrain image close to cloudy image as Adversarial Cloud. Cloud is natural and common in remote sensing images, however, camouflaging cloud based adversarial attack and defense for remote sensing images are not well studied before. Furthermore, we design DefenseNet as a learn-able pre-processing to the adversarial cloudy images so as to preserve the performance of the deep learning based remote sensing SOD model, without tuning the already deployed deep SOD model. By considering both regular and generalized adversarial examples, the proposed DefenseNet can defend the proposed Adversarial Cloud in white-box setting and other attack methods in black-box setting. Experimental results on a synthesized benchmark from the public remote sensing SOD dataset (EORSSD) show the promising defense against adversarial cloud attacks.

Abstract: 3D single object tracking with point clouds is a critical task in 3D computer vision. Previous methods usually input the last two frames and use the predicted box to get the template point cloud in previous frame and the search area point cloud in the current frame respectively, then use similarity-based or motion-based methods to predict the current box. Although these methods achieved good tracking performance, they ignore the historical information of the target, which is important for tracking. In this paper, compared to inputting two frames of point clouds, we input multi-frame of point clouds to encode the spatio-temporal information of the target and learn the motion information of the target implicitly, which could build the correlations among different frames to track the target in the current frame efficiently. Meanwhile, rather than directly using the point feature for feature fusion, we first crop the point cloud features into many patches and then use sparse attention mechanism to encode the patch-level similarity and finally fuse the multi-frame features. Extensive experiments show that our method achieves competitive results on challenging large-scale benchmarks (62.6% in KITTI and 49.66% in NuScenes).

Abstract: The success of a deep neural network (DNN) heavily relies on the details of the training scheme; e.g., training data, architectures, hyper-parameters, etc. Recent backdoor attacks suggest that an adversary can take advantage of such training details and compromise the integrity of a DNN. Our studies show that a backdoor model is usually optimized to a bad local minima, i.e. sharper minima as compared to a benign model. Intuitively, a backdoor model can be purified by reoptimizing the model to a smoother minima through fine-tuning with a few clean validation data. However, fine-tuning all DNN parameters often requires huge computational costs and often results in sub-par clean test performance. To address this concern, we propose a novel backdoor purification technique, Natural Gradient Fine-tuning (NGF), which focuses on removing the backdoor by fine-tuning only one layer. Specifically, NGF utilizes a loss surface geometry-aware optimizer that can successfully overcome the challenge of reaching a smooth minima under a one-layer optimization scenario. To enhance the generalization performance of our proposed method, we introduce a clean data distribution-aware regularizer based on the knowledge of loss surface curvature matrix, i.e., Fisher Information Matrix. Extensive experiments show that the proposed method achieves state-of-the-art performance on a wide range of backdoor defense benchmarks: four different datasets- CIFAR10, GTSRB, Tiny-ImageNet, and ImageNet; 13 recent backdoor attacks, e.g. Blend, Dynamic, WaNet, ISSBA, etc.

Abstract: Deep neural networks (DNNs) offer the highest performance in a wide range of applications in computer vision. These results rely on over-parameterized backbones, which are expensive to run. This computational burden can be dramatically reduced by quantizing (in either data-free (DFQ), post-training (PTQ) or quantization-aware training (QAT) scenarios) floating point values to ternary values (2 bits, with each weight taking value in {-1,0,1}). In this context, we observe that rounding to nearest minimizes the expected error given a uniform distribution and thus does not account for the skewness and kurtosis of the weight distribution, which strongly affects ternary quantization performance. This raises the following question: shall one minimize the highest or average quantization error? To answer this, we design two operators: TQuant and MQuant that correspond to these respective minimization tasks. We show experimentally that our approach allows to significantly improve the performance of ternary quantization through a variety of scenarios in DFQ, PTQ and QAT and give strong insights to pave the way for future research in deep neural network quantization.

Abstract: Depth perception is a crucial component of monoc-ular 3D detection tasks that typically involve ill-posed problems. In light of the success of sample mining techniques in 2D object detection, we propose a simple yet effective mining strategy for improving depth perception in 3D object detection. Concretely, we introduce a plain metric to evaluate the quality of depth predictions, which chooses the mined sample for the model. Moreover, we propose a Gradient-aware and Model-perceive Mining strategy (GMM) for depth learning, which exploits the predicted depth quality for better depth learning through easy mining. GMM is a general strategy that can be readily applied to several state-of-the-art monocular 3D detectors, improving the accuracy of depth prediction. Extensive experiments on the nuScenes dataset demonstrate that the proposed methods significantly improve the performance of 3D object detection while outperforming other state-of-the-art sample mining techniques by a considerable margin. On the nuScenes benchmark, GMM achieved the state-of-the-art (42.1% mAP and 47.3% NDS) performance in monocular object detection.

Abstract: We present CausalVLR (Causal Visual-Linguistic Reasoning), an open-source toolbox containing a rich set of state-of-the-art causal relation discovery and causal inference methods for various visual-linguistic reasoning tasks, such as VQA, image/video captioning, medical report generation, model generalization and robustness, etc. These methods have been included in the toolbox with PyTorch implementations under NVIDIA computing system. It not only includes training and inference codes, but also provides model weights. We believe this toolbox is by far the most complete visual-linguitic causal reasoning toolbox. We wish that the toolbox and benchmark could serve the growing research community by providing a flexible toolkit to re-implement existing methods and develop their own new causal reasoning methods. Code and models are available at https://github.com/HCPLab-SYSU/Causal-VLReasoning. The project is under active development by HCP-Lab's contributors and we will keep this document updated.

Abstract: The expanding market for e-comics has spurred interest in the development of automated methods to analyze comics. For further understanding of comics, an automated approach is needed to link text in comics to characters speaking the words. Comics speaker detection research has practical applications, such as automatic character assignment for audiobooks, automatic translation according to characters' personalities, and inference of character relationships and stories. To deal with the problem of insufficient speaker-to-text annotations, we created a new annotation dataset Manga109Dialog based on Manga109. Manga109Dialog is the world's largest comics speaker annotation dataset, containing 132,692 speaker-to-text pairs. We further divided our dataset into different levels by prediction difficulties to evaluate speaker detection methods more appropriately. Unlike existing methods mainly based on distances, we propose a deep learning-based method using scene graph generation models. Due to the unique features of comics, we enhance the performance of our proposed model by considering the frame reading order. We conducted experiments using Manga109Dialog and other datasets. Experimental results demonstrate that our scene-graph-based approach outperforms existing methods, achieving a prediction accuracy of over 75%.

Abstract: Object detection and segmentation are two core modules of an autonomous vehicle perception system. They should have high efficiency and low latency while reducing computational complexity. Currently, the most commonly used algorithms are based on deep neural networks, which guarantee high efficiency but require high-performance computing platforms. In the case of autonomous vehicles, i.e. cars, but also drones, it is necessary to use embedded platforms with limited computing power, which makes it difficult to meet the requirements described above. A reduction in the complexity of the network can be achieved by using an appropriate: architecture, representation (reduced numerical precision, quantisation, pruning), and computing platform. In this paper, we focus on the first factor - the use of so-called detection-segmentation networks as a component of a perception system. We considered the task of segmenting the drivable area and road markings in combination with the detection of selected objects (pedestrians, traffic lights, and obstacles). We compared the performance of three different architectures described in the literature: MultiTask V3, HybridNets, and YOLOP. We conducted the experiments on a custom dataset consisting of approximately 500 images of the drivable area and lane markings, and 250 images of detected objects. Of the three methods analysed, MultiTask V3 proved to be the best, achieving 99% mAP_50 for detection, 97% MIoU for drivable area segmentation, and 91% MIoU for lane segmentation, as well as 124 fps on the RTX 3060 graphics card. This architecture is a good solution for embedded perception systems for autonomous vehicles. The code is available at: https://github.com/vision-agh/MMAR_2023.

Abstract: Can knowing where you are assist in perceiving objects in your surroundings, especially under adverse weather and lighting conditions? In this work we investigate whether a prior map can be leveraged to aid in the detection of dynamic objects in a scene without the need for a 3D map or pixel-level map-query correspondences. We contribute an algorithm which refines an initial set of candidate object detections and produces a refined subset of highly accurate detections using a prior map. We begin by using visual place recognition (VPR) to retrieve a reference map image for a given query image, then use a binary classification neural network that compares the query and mapping image regions to validate the query detection. Once our classification network is trained, on approximately 1000 query-map image pairs, it is able to improve the performance of vehicle detection when combined with an existing off-the-shelf vehicle detector. We demonstrate our approach using standard datasets across two cities (Oxford and Zurich) under different settings of train-test separation of map-query traverse pairs. We further emphasize the performance gains of our approach against alternative design choices and show that VPR suffices for the task, eliminating the need for precise ground truth localization.

Abstract: Object detection in 3D is a crucial aspect in the context of autonomous vehicles and drones. However, prototyping detection algorithms is time-consuming and costly in terms of energy and environmental impact. To address these challenges, one can check the effectiveness of different models by training on a subset of the original training set. In this paper, we present a comparison of three algorithms for selecting such a subset - random sampling, random per class sampling, and our proposed MONSPeC (Maximum Object Number Sampling per Class). We provide empirical evidence for the superior effectiveness of random per class sampling and MONSPeC over basic random sampling. By replacing random sampling with one of the more efficient algorithms, the results obtained on the subset are more likely to transfer to the results on the entire dataset. The code is available at: https://github.com/vision-agh/monspec.

Abstract: It is an open secret that ImageNet is treated as the panacea of pretraining. Particularly in medical machine learning, models not trained from scratch are often finetuned based on ImageNet-pretrained models. We posit that pretraining on data from the domain of the downstream task should almost always be preferred instead. We leverage RadNet-12M, a dataset containing more than 12 million computed tomography (CT) image slices, to explore the efficacy of self-supervised pretraining on medical and natural images. Our experiments cover intra- and cross-domain transfer scenarios, varying data scales, finetuning vs. linear evaluation, and feature space analysis. We observe that intra-domain transfer compares favorably to cross-domain transfer, achieving comparable or improved performance (0.44% - 2.07% performance increase using RadNet pretraining, depending on the experiment) and demonstrate the existence of a domain boundary-related generalization gap and domain-specific learned features.

Abstract: Isolated Sign Language Recognition (SLR) has mostly been applied on relatively large datasets containing signs executed slowly and clearly by a limited group of signers. In real-world scenarios, however, we are met with challenging visual conditions, coarticulated signing, small datasets, and the need for signer independent models. To tackle this difficult problem, we require a robust feature extractor to process the sign language videos. One could expect human pose estimators to be ideal candidates. However, due to a domain mismatch with their training sets and challenging poses in sign language, they lack robustness on sign language data and image based models often still outperform keypoint based models. Furthermore, whereas the common practice of transfer learning with image based models yields even higher accuracy, keypoint based models are typically trained from scratch on every SLR dataset. These factors limit their usefulness for SLR. From the existing literature, it is also not clear which, if any, pose estimator performs best for SLR. We compare the three most popular pose estimators for SLR: OpenPose, MMPose and MediaPipe. We show that through keypoint normalization, missing keypoint imputation, and learning a pose embedding, we can obtain significantly better results and enable transfer learning. We show that keypoint-based embeddings contain cross-lingual features: they can transfer between sign languages and achieve competitive performance even when fine-tuning only the classifier layer of an SLR model on a target sign language. We furthermore achieve better performance using fine-tuned transferred embeddings than models trained only on the target sign language. The application of these embeddings could prove particularly useful for low resource sign languages in the future.

Abstract: Recently, the quality and performance of text-to-image generation significantly advanced due to the impressive results of diffusion models. However, text-to-image diffusion models still fail to generate high fidelity content with respect to the input prompt. One problem where text-to-diffusion models struggle is generating the exact number of objects specified in the text prompt. E.g. given a prompt "five apples and ten lemons on a table", diffusion-generated images usually contain the wrong number of objects. In this paper, we propose a method to improve diffusion models to focus on producing the correct object count given the input prompt. We adopt a counting network that performs reference-less class-agnostic counting for any given image. We calculate the gradients of the counting network and refine the predicted noise for each step. To handle multiple types of objects in the prompt, we use novel attention map guidance to obtain high-fidelity masks for each object. Finally, we guide the denoising process by the calculated gradients for each object. Through extensive experiments and evaluation, we demonstrate that our proposed guidance method greatly improves the fidelity of diffusion models to object count.

Abstract: Recent advancements in technology have expanded the possibilities of human action recognition by leveraging 3D data, which offers a richer representation of actions through the inclusion of depth information, enabling more accurate analysis of spatial and temporal characteristics. However, 3D human action recognition is a challenging task due to the irregularity and Disarrangement of the data points in action sequences. In this context, we present our novel model for human action recognition from fixed topology mesh sequences based on Spiral Auto-encoder and Transformer Network, namely SpATr. The proposed method first disentangles space and time in the mesh sequences. Then, an auto-encoder is utilized to extract spatial geometrical features, and tiny transformer is used to capture the temporal evolution of the sequence. Previous methods either use 2D depth images, sample skeletons points or they require a huge amount of memory leading to the ability to process short sequences only. In this work, we show competitive recognition rate and high memory efficiency by building our auto-encoder based on spiral convolutions, which are light weight convolution directly applied to mesh data with fixed topologies, and by modeling temporal evolution using a attention, that can handle large sequences. The proposed method is evaluated on on two 3D human action datasets: MoVi and BMLrub from the Archive of Motion Capture As Surface Shapes (AMASS). The results analysis shows the effectiveness of our method in 3D human action recognition while maintaining high memory efficiency. The code will soon be made publicly available.

Abstract: Magnitude pruning is one of the mainstream methods in lightweight architecture design whose goal is to extract subnetworks with the largest weight connections. This method is known to be successful, but under very high pruning regimes, it suffers from topological inconsistency which renders the extracted subnetworks disconnected, and this hinders their generalization ability. In this paper, we devise a novel magnitude pruning method that allows extracting subnetworks while guarantying their topological consistency. The latter ensures that only accessible and co-accessible -- impactful -- connections are kept in the resulting lightweight networks. Our solution is based on a novel reparametrization and two supervisory bi-directional networks which implement accessibility/co-accessibility and guarantee that only connected subnetworks will be selected during training. This solution allows enhancing generalization significantly, under very high pruning regimes, as corroborated through extensive experiments, involving graph convolutional networks, on the challenging task of skeleton-based action recognition.

Abstract: Burst super-resolution (BurstSR) aims at reconstructing a high-resolution (HR) image from a sequence of low-resolution (LR) and noisy images, which is conducive to enhancing the imaging effects of smartphones with limited sensors. The main challenge of BurstSR is to effectively combine the complementary information from input frames, while existing methods still struggle with it. In this paper, we suggest fusing cues frame-by-frame with an efficient and flexible recurrent network. In particular, we emphasize the role of the base-frame and utilize it as a key prompt to guide the knowledge acquisition from other frames in every recurrence. Moreover, we introduce an implicit weighting loss to improve the model's flexibility in facing input frames with variable numbers. Extensive experiments on both synthetic and real-world datasets demonstrate that our method achieves better results than state-of-the-art ones. Codes and pre-trained models are available at https://github.com/ZcsrenlongZ/RBSR.

Abstract: The event streams generated by dynamic vision sensors (DVS) are sparse and non-uniform in the spatial domain, while still dense and redundant in the temporal domain. Although spiking neural network (SNN), the event-driven neuromorphic model, has the potential to extract spatio-temporal features from the event streams, it is not effective and efficient. Based on the above, we propose an events sparsification spiking framework dubbed as Razor SNN, pruning pointless event frames progressively. Concretely, we extend the dynamic mechanism based on the global temporal embeddings, reconstruct the features, and emphasize the events effect adaptively at the training stage. During the inference stage, eliminate fruitless frames hierarchically according to a binary mask generated by the trained temporal embeddings. Comprehensive experiments demonstrate that our Razor SNN achieves competitive performance consistently on four events-based benchmarks: DVS 128 Gesture, N-Caltech 101, CIFAR10-DVS and SHD.

Abstract: Following the tracking-by-attention paradigm, this paper introduces an object-centric, transformer-based framework for tracking in 3D. Traditional model-based tracking approaches incorporate the geometric effect of object- and ego motion between frames with a geometric motion model. Inspired by this, we propose S.T.A.R.-Track, which uses a novel latent motion model (LMM) to additionally adjust object queries to account for changes in viewing direction and lighting conditions directly in the latent space, while still modeling the geometric motion explicitly. Combined with a novel learnable track embedding that aids in modeling the existence probability of tracks, this results in a generic tracking framework that can be integrated with any query-based detector. Extensive experiments on the nuScenes benchmark demonstrate the benefits of our approach, showing state-of-the-art performance for DETR3D-based trackers while drastically reducing the number of identity switches of tracks at the same time.

Abstract: Indirect time-of-flight (iToF) imaging allows us to capture dense depth information at a low cost. However, iToF imaging often suffers from multipath interference (MPI) artifacts in the presence of scattering media, resulting in severe depth-accuracy degradation. For instance, iToF cameras cannot measure depth accurately through fog because ToF active illumination scatters back to the sensor before reaching the farther target surface. In this work, we propose a polarimetric iToF imaging method that can capture depth information robustly through scattering media. Our observations on the principle of indirect ToF imaging and polarization of light allow us to formulate a novel computational model of scattering-aware polarimetric phase measurements that enables us to correct MPI errors. We first devise a scattering-aware polarimetric iToF model that can estimate the phase of unpolarized backscattered light. We then combine the optical filtering of polarization and our computational modeling of unpolarized backscattered light via scattering analysis of phase and amplitude. This allows us to tackle the MPI problem by estimating the scattering energy through the participating media. We validate our method on an experimental setup using a customized off-the-shelf iToF camera. Our method outperforms baseline methods by a significant margin by means of our scattering model and polarimetric phase measurements.

Abstract: Generating learning-friendly representations for points in space is a fundamental and long-standing problem in ML. Recently, multi-scale encoding schemes (such as Space2Vec and NeRF) were proposed to directly encode any point in 2D/3D Euclidean space as a high-dimensional vector, and has been successfully applied to various geospatial prediction and generative tasks. However, all current 2D and 3D location encoders are designed to model point distances in Euclidean space. So when applied to large-scale real-world GPS coordinate datasets, which require distance metric learning on the spherical surface, both types of models can fail due to the map projection distortion problem (2D) and the spherical-to-Euclidean distance approximation error (3D). To solve these problems, we propose a multi-scale location encoder called Sphere2Vec which can preserve spherical distances when encoding point coordinates on a spherical surface. We developed a unified view of distance-reserving encoding on spheres based on the DFS. We also provide theoretical proof that the Sphere2Vec preserves the spherical surface distance between any two points, while existing encoding schemes do not. Experiments on 20 synthetic datasets show that Sphere2Vec can outperform all baseline models on all these datasets with up to 30.8% error rate reduction. We then apply Sphere2Vec to three geo-aware image classification tasks - fine-grained species recognition, Flickr image recognition, and remote sensing image classification. Results on 7 real-world datasets show the superiority of Sphere2Vec over multiple location encoders on all three tasks. Further analysis shows that Sphere2Vec outperforms other location encoder models, especially in the polar regions and data-sparse areas because of its nature for spherical surface distance preservation. Code and data are available at https://gengchenmai.github.io/sphere2vec-website/.

Abstract: Depth is a very important modality in computer vision, typically used as complementary information to RGB, provided by RGB-D cameras. In this work, we show that it is possible to obtain the same level of accuracy as RGB-D cameras on a semantic segmentation task using infrared (IR) and depth images from a single Time-of-Flight (ToF) camera. In order to fuse the IR and depth modalities of the ToF camera, we introduce a method utilizing depth-specific convolutions in a multi-task learning framework. In our evaluation on an in-car segmentation dataset, we demonstrate the competitiveness of our method against the more costly RGB-D approaches.

Abstract: This paper tackles the problem of object counting in images. Existing approaches rely on extensive training data with point annotations for each object, making data collection labor-intensive and time-consuming. To overcome this, we propose a training-free object counter that treats the counting task as a segmentation problem. Our approach leverages the Segment Anything Model (SAM), known for its high-quality masks and zero-shot segmentation capability. However, the vanilla mask generation method of SAM lacks class-specific information in the masks, resulting in inferior counting accuracy. To overcome this limitation, we introduce a prior-guided mask generation method that incorporates three types of priors into the segmentation process, enhancing efficiency and accuracy. Additionally, we tackle the issue of counting objects specified through free-form text by proposing a two-stage approach that combines reference object selection and prior-guided mask generation. Extensive experiments on standard datasets demonstrate the competitive performance of our training-free counter compared to learning-based approaches. This paper presents a promising solution for counting objects in various scenarios without the need for extensive data collection and model training. Code is available at https://github.com/shizenglin/training-free-object-counter.

Abstract: Neural 3D scene reconstruction methods have achieved impressive performance when reconstructing complex geometry and low-textured regions in indoor scenes. However, these methods heavily rely on 3D data which is costly and time-consuming to obtain in real world. In this paper, we propose a novel neural reconstruction method that reconstructs scenes using sparse depth under the plane constraints without 3D supervision. We introduce a signed distance function field, a color field, and a probability field to represent a scene. We optimize these fields to reconstruct the scene by using differentiable ray marching with accessible 2D images as supervision. We improve the reconstruction quality of complex geometry scene regions with sparse depth obtained by using the geometric constraints. The geometric constraints project 3D points on the surface to similar-looking regions with similar features in different 2D images. We impose the plane constraints to make large planes parallel or vertical to the indoor floor. Both two constraints help reconstruct accurate and smooth geometry structures of the scene. Without 3D supervision, our method achieves competitive performance compared with existing methods that use 3D supervision on the ScanNet dataset.

Abstract: From an image of a person, we can easily infer the natural 3D pose and shape of the person even if ambiguity exists. This is because we have a mental model that allows us to imagine a person's appearance at different viewing directions from a given image and utilize the consistency between them for inference. However, existing human mesh recovery methods only consider the direction in which the image was taken due to their structural limitations. Hence, we propose "Implicit 3D Human Mesh Recovery (ImpHMR)" that can implicitly imagine a person in 3D space at the feature-level via Neural Feature Fields. In ImpHMR, feature fields are generated by CNN-based image encoder for a given image. Then, the 2D feature map is volume-rendered from the feature field for a given viewing direction, and the pose and shape parameters are regressed from the feature. To utilize consistency with pose and shape from unseen-view, if there are 3D labels, the model predicts results including the silhouette from an arbitrary direction and makes it equal to the rotated ground-truth. In the case of only 2D labels, we perform self-supervised learning through the constraint that the pose and shape parameters inferred from different directions should be the same. Extensive evaluations show the efficacy of the proposed method.

Abstract: Large-scale visual-language pre-trained models (VLPM) have proven their excellent performance in downstream object detection for natural scenes. However, zero-shot nuclei detection on H\&E images via VLPMs remains underexplored. The large gap between medical images and the web-originated text-image pairs used for pre-training makes it a challenging task. In this paper, we attempt to explore the potential of the object-level VLPM, Grounded Language-Image Pre-training (GLIP) model, for zero-shot nuclei detection. Concretely, an automatic prompts design pipeline is devised based on the association binding trait of VLPM and the image-to-text VLPM BLIP, avoiding empirical manual prompts engineering. We further establish a self-training framework, using the automatically designed prompts to generate the preliminary results as pseudo labels from GLIP and refine the predicted boxes in an iterative manner. Our method achieves a remarkable performance for label-free nuclei detection, surpassing other comparison methods. Foremost, our work demonstrates that the VLPM pre-trained on natural image-text pairs exhibits astonishing potential for downstream tasks in the medical field as well. Code will be released at https://github.com/wuyongjianCODE/VLPMNuD.

Abstract: Recent years have witnessed impressive results of pre-trained vision-language models on knowledge-intensive tasks such as visual question answering (VQA). Despite the recent advances in VQA, existing methods mainly adopt a discriminative formulation that predicts answers within a pre-defined label set, leading to easy overfitting on low-resource domains with limited labeled data (e.g., medicine) and poor generalization under domain shift to another dataset. To tackle this limitation, we propose a novel generative model enhanced by multimodal prompt retrieval (MPR) that integrates retrieved prompts and multimodal features to generate answers in free text. Our generative model enables rapid zero-shot dataset adaptation to unseen data distributions and open-set answer labels across datasets. Our experiments on medical VQA tasks show that MPR outperforms its non-retrieval counterpart by up to 30% accuracy points in a few-shot domain adaptation setting.

Abstract: In this paper, we address a complex but practical scenario in semi-supervised learning (SSL) named open-set SSL, where unlabeled data contain both in-distribution (ID) and out-of-distribution (OOD) samples. Unlike previous methods that only consider ID samples to be useful and aim to filter out OOD ones completely during training, we argue that the exploration and exploitation of both ID and OOD samples can benefit SSL. To support our claim, i) we propose a prototype-based clustering and identification algorithm that explores the inherent similarity and difference among samples at feature level and effectively cluster them around several predefined ID and OOD prototypes, thereby enhancing feature learning and facilitating ID/OOD identification; ii) we propose an importance-based sampling method that exploits the difference in importance of each ID and OOD sample to SSL, thereby reducing the sampling bias and improving the training. Our proposed method achieves state-of-the-art in several challenging benchmarks, and improves upon existing SSL methods even when ID samples are totally absent in unlabeled data.

Abstract: Neural Radiance Field (NeRF) has been a mainstream in novel view synthesis with its remarkable quality of rendered images and simple architecture. Although NeRF has been developed in various directions improving continuously its performance, the necessity of a dense set of multi-view images still exists as a stumbling block to progress for practical application. In this work, we propose FlipNeRF, a novel regularization method for few-shot novel view synthesis by utilizing our proposed flipped reflection rays. The flipped reflection rays are explicitly derived from the input ray directions and estimated normal vectors, and play a role of effective additional training rays while enabling to estimate more accurate surface normals and learn the 3D geometry effectively. Since the surface normal and the scene depth are both derived from the estimated densities along a ray, the accurate surface normal leads to more exact depth estimation, which is a key factor for few-shot novel view synthesis. Furthermore, with our proposed Uncertainty-aware Emptiness Loss and Bottleneck Feature Consistency Loss, FlipNeRF is able to estimate more reliable outputs with reducing floating artifacts effectively across the different scene structures, and enhance the feature-level consistency between the pair of the rays cast toward the photo-consistent pixels without any additional feature extractor, respectively. Our FlipNeRF achieves the SOTA performance on the multiple benchmarks across all the scenarios.

Abstract: Motion prediction is crucial for autonomous driving systems to understand complex driving scenarios and make informed decisions. However, this task is challenging due to the diverse behaviors of traffic participants and complex environmental contexts. In this paper, we propose Motion TRansformer (MTR) frameworks to address these challenges. The initial MTR framework utilizes a transformer encoder-decoder structure with learnable intention queries, enabling efficient and accurate prediction of future trajectories. By customizing intention queries for distinct motion modalities, MTR improves multimodal motion prediction while reducing reliance on dense goal candidates. The framework comprises two essential processes: global intention localization, identifying the agent's intent to enhance overall efficiency, and local movement refinement, adaptively refining predicted trajectories for improved accuracy. Moreover, we introduce an advanced MTR++ framework, extending the capability of MTR to simultaneously predict multimodal motion for multiple agents. MTR++ incorporates symmetric context modeling and mutually-guided intention querying modules to facilitate future behavior interaction among multiple agents, resulting in scene-compliant future trajectories. Extensive experimental results demonstrate that the MTR framework achieves state-of-the-art performance on the highly-competitive motion prediction benchmarks, while the MTR++ framework surpasses its precursor, exhibiting enhanced performance and efficiency in predicting accurate multimodal future trajectories for multiple agents.

Abstract: Large language models have recently shown human level performance on a variety of reasoning tasks. However, the ability of these models to perform complex visual reasoning has not been studied in detail yet. A key challenge in many visual reasoning tasks is that the visual information needs to be tightly integrated in the reasoning process. We propose to address this challenge by drawing inspiration from human visual problem solving which depends on a variety of low-level visual capabilities. It can often be cast as the three step-process of ``Look, Remember, Reason'': visual information is incrementally extracted using low-level visual routines in a step-by-step fashion until a final answer is reached. We follow the same paradigm to enable existing large language models, with minimal changes to the architecture, to solve visual reasoning problems. To this end, we introduce rationales over the visual input that allow us to integrate low-level visual capabilities, such as object recognition and tracking, as surrogate tasks. We show competitive performance on diverse visual reasoning tasks from the CLEVR, CATER, and ACRE datasets over state-of-the-art models designed specifically for these tasks.

Abstract: Generative AI has made significant strides in computer vision, particularly in image/video synthesis conditioned on text descriptions. Despite the advancements, it remains challenging especially in the generation of human-centric content such as dance synthesis. Existing dance synthesis methods struggle with the gap between synthesized content and real-world dance scenarios. In this paper, we define a new problem setting: Referring Human Dance Generation, which focuses on real-world dance scenarios with three important properties: (i) Faithfulness: the synthesis should retain the appearance of both human subject foreground and background from the reference image, and precisely follow the target pose; (ii) Generalizability: the model should generalize to unseen human subjects, backgrounds, and poses; (iii) Compositionality: it should allow for composition of seen/unseen subjects, backgrounds, and poses from different sources. To address these challenges, we introduce a novel approach, DISCO, which includes a novel model architecture with disentangled control to improve the faithfulness and compositionality of dance synthesis, and an effective human attribute pre-training for better generalizability to unseen humans. Extensive qualitative and quantitative results demonstrate that DISCO can generate high-quality human dance images and videos with diverse appearances and flexible motions. Code, demo, video and visualization are available at: https://disco-dance.github.io/.

Abstract: Federated learning (FL) has gained significant traction as a privacy-preserving algorithm, but the underlying resembles of federated learning algorithm like Federated averaging (FED Avg) or Federated SGD (FED SGD) to ensemble learning algorithms has not been fully explored. The purpose of this paper is to examine the application of FL to object detection as a method to enhance generalizability, and to compare its performance against a centralized training approach for an object detection algorithm. Specifically, we investigate the performance of a YOLOv5 model trained using FL across multiple clients and employ a random sampling strategy without replacement, so each client holds a portion of the same dataset used for centralized training. Our experimental results showcase the superior efficiency of the FL object detector's global model in generating accurate bounding boxes for unseen objects, with the test set being a mixture of objects from two distinct clients not represented in the training dataset. These findings suggest that FL can be viewed from an ensemble algorithm perspective, akin to a synergistic blend of Bagging and Boosting techniques. As a result, FL can be seen not only as a method to enhance privacy, but also as a method to enhance the performance of a machine learning model.

Abstract: In this work, we introduce Semantic Pyramid AutoEncoder (SPAE) for enabling frozen LLMs to perform both understanding and generation tasks involving non-linguistic modalities such as images or videos. SPAE converts between raw pixels and interpretable lexical tokens (or words) extracted from the LLM's vocabulary. The resulting tokens capture both the semantic meaning and the fine-grained details needed for visual reconstruction, effectively translating the visual content into a language comprehensible to the LLM, and empowering it to perform a wide array of multimodal tasks. Our approach is validated through in-context learning experiments with frozen PaLM 2 and GPT 3.5 on a diverse set of image understanding and generation tasks. Our method marks the first successful attempt to enable a frozen LLM to generate image content while surpassing state-of-the-art performance in image understanding tasks, under the same setting, by over 25%.

Abstract: We present Magic123, a two-stage coarse-to-fine approach for high-quality, textured 3D meshes generation from a single unposed image in the wild using both2D and 3D priors. In the first stage, we optimize a neural radiance field to produce a coarse geometry. In the second stage, we adopt a memory-efficient differentiable mesh representation to yield a high-resolution mesh with a visually appealing texture. In both stages, the 3D content is learned through reference view supervision and novel views guided by a combination of 2D and 3D diffusion priors. We introduce a single trade-off parameter between the 2D and 3D priors to control exploration (more imaginative) and exploitation (more precise) of the generated geometry. Additionally, we employ textual inversion and monocular depth regularization to encourage consistent appearances across views and to prevent degenerate solutions, respectively. Magic123 demonstrates a significant improvement over previous image-to-3D techniques, as validated through extensive experiments on synthetic benchmarks and diverse real-world images. Our code, models, and generated 3D assets are available at https://github.com/guochengqian/Magic123.

Abstract: Vision transformers (ViTs) have significantly changed the computer vision landscape and have periodically exhibited superior performance in vision tasks compared to convolutional neural networks (CNNs). Although the jury is still out on which model type is superior, each has unique inductive biases that shape their learning and generalization performance. For example, ViTs have interesting properties with respect to early layer non-local feature dependence, as well as self-attention mechanisms which enhance learning flexibility, enabling them to ignore out-of-context image information more effectively. We hypothesize that this power to ignore out-of-context information (which we name $\textit{patch selectivity}$), while integrating in-context information in a non-local manner in early layers, allows ViTs to more easily handle occlusion. In this study, our aim is to see whether we can have CNNs $\textit{simulate}$ this ability of patch selectivity by effectively hardwiring this inductive bias using Patch Mixing data augmentation, which consists of inserting patches from another image onto a training image and interpolating labels between the two image classes. Specifically, we use Patch Mixing to train state-of-the-art ViTs and CNNs, assessing its impact on their ability to ignore out-of-context patches and handle natural occlusions. We find that ViTs do not improve nor degrade when trained using Patch Mixing, but CNNs acquire new capabilities to ignore out-of-context information and improve on occlusion benchmarks, leaving us to conclude that this training method is a way of simulating in CNNs the abilities that ViTs already possess. We will release our Patch Mixing implementation and proposed datasets for public use. Project page: https://arielnlee.github.io/PatchMixing/

Abstract: In her influential 1988 paper, Situated Knowledges, Donna Haraway uses vision and perspective as a metaphor to discuss scientific knowledge. Today, egocentric computer vision discusses many of the same issues, except in a literal vision context. In this short position paper, we collapse that metaphor, and explore the interactions between feminist epistemology and egocentric CV as "Egocentric Epistemology." Using this framework, we argue for the use of qualitative, human-centric methods as a complement to performance benchmarks, to center both the literal and metaphorical perspective of human crowd workers in CV.

Abstract: In this paper, a parts based loss is considered for finetune registering knee joint areas. Here the parts are defined as abstract feature vectors with location and they are automatically selected from a reference image. For a test image the detected parts are encouraged to have a similar spatial configuration than the corresponding parts in the reference image.

Abstract: Recently, CLIP-based approaches have exhibited remarkable performance on generalization and few-shot learning tasks, fueled by the power of contrastive language-vision pre-training. In particular, prompt tuning has emerged as an effective strategy to adapt the pre-trained language-vision models to downstream tasks by employing task-related textual tokens. Motivated by this progress, in this work we question whether other fundamental problems, such as weakly supervised semantic segmentation (WSSS), can benefit from prompt tuning. Our findings reveal two interesting observations that shed light on the impact of prompt tuning on WSSS. First, modifying only the class token of the text prompt results in a greater impact on the Class Activation Map (CAM), compared to arguably more complex strategies that optimize the context. And second, the class token associated with the image ground truth does not necessarily correspond to the category that yields the best CAM. Motivated by these observations, we introduce a novel approach based on a PrOmpt cLass lEarning (POLE) strategy. Through extensive experiments we demonstrate that our simple, yet efficient approach achieves SOTA performance in a well-known WSSS benchmark. These results highlight not only the benefits of language-vision models in WSSS but also the potential of prompt learning for this problem. The code is available at https://github.com/rB080/WSS_POLE.

Abstract: This research paper addresses the challenge of detecting obscured wildfires (when the fire flames are covered by trees, smoke, clouds, and other natural barriers) in real-time using drones equipped only with RGB cameras. We propose a novel methodology that employs semantic segmentation based on the temporal analysis of smoke patterns in video sequences. Our approach utilizes an encoder-decoder architecture based on deep convolutional neural network architecture with a pre-trained CNN encoder and 3D convolutions for decoding while using sequential stacking of features to exploit temporal variations. The predicted fire locations can assist drones in effectively combating forest fires and pinpoint fire retardant chemical drop on exact flame locations. We applied our method to a curated dataset derived from the FLAME2 dataset that includes RGB video along with IR video to determine the ground truth. Our proposed method has a unique property of detecting obscured fire and achieves a Dice score of 85.88%, while achieving a high precision of 92.47% and classification accuracy of 90.67% on test data showing promising results when inspected visually. Indeed, our method outperforms other methods by a significant margin in terms of video-level fire classification as we obtained about 100% accuracy using MobileNet+CBAM as the encoder backbone.

Abstract: This paper presents a novel generative model for Computer Aided Design (CAD) that 1) represents high-level design concepts of a CAD model as a three-level hierarchical tree of neural codes, from global part arrangement down to local curve geometry; and 2) controls the generation or completion of CAD models by specifying the target design using a code tree. Concretely, a novel variant of a vector quantized VAE with "masked skip connection" extracts design variations as neural codebooks at three levels. Two-stage cascaded auto-regressive transformers learn to generate code trees from incomplete CAD models and then complete CAD models following the intended design. Extensive experiments demonstrate superior performance on conventional tasks such as random generation while enabling novel interaction capabilities on conditional generation tasks. The code is available at https://github.com/samxuxiang/hnc-cad.

Abstract: Large pretrained plain vision Transformers (ViTs) have been the workhorse for many downstream tasks. However, existing works utilizing off-the-shelf ViTs are inefficient in terms of training and deployment, because adopting ViTs with individual sizes requires separate training and is restricted by fixed performance-efficiency trade-offs. In this paper, we are inspired by stitchable neural networks, which is a new framework that cheaply produces a single model that covers rich subnetworks by stitching pretrained model families, supporting diverse performance-efficiency trade-offs at runtime. Building upon this foundation, we introduce SN-Netv2, a systematically improved model stitching framework to facilitate downstream task adaptation. Specifically, we first propose a Two-way stitching scheme to enlarge the stitching space. We then design a resource-constrained sampling strategy that takes into account the underlying FLOPs distributions in the space for improved sampling. Finally, we observe that learning stitching layers is a low-rank update, which plays an essential role on downstream tasks to stabilize training and ensure a good Pareto frontier. With extensive experiments on ImageNet-1K, ADE20K, COCO-Stuff-10K, NYUv2 and COCO-2017, SN-Netv2 demonstrates strong ability to serve as a flexible vision backbone, achieving great advantages in both training efficiency and adaptation. Code will be released at https://github.com/ziplab/SN-Netv2.