Computer Vision and Pattern Recognition (cs.CV)

Abstract: Trained computer vision models are assumed to solve vision tasks by imitating human behavior learned from training labels. Most efforts in recent vision research focus on measuring the model task performance using standardized benchmarks. Limited work has been done to understand the perceptual difference between humans and machines. To fill this gap, our study first quantifies and analyzes the statistical distributions of mistakes from the two sources. We then explore human vs. machine expertise after ranking tasks by difficulty levels. Even when humans and machines have similar overall accuracies, the distribution of answers may vary. Leveraging the perceptual difference between humans and machines, we empirically demonstrate a post-hoc human-machine collaboration that outperforms humans or machines alone.

Abstract: Vision Transformers have shown great performance in single tasks such as classification and segmentation. However, real-world problems are not isolated, which calls for vision transformers that can perform multiple tasks concurrently. Existing multi-task vision transformers are handcrafted and heavily rely on human expertise. In this work, we propose a novel one-shot neural architecture search framework, dubbed AutoTaskFormer (Automated Multi-Task Vision TransFormer), to automate this process. AutoTaskFormer not only identifies the weights to share across multiple tasks automatically, but also provides thousands of well-trained vision transformers with a wide range of parameters (e.g., number of heads and network depth) for deployment under various resource constraints. Experiments on both small-scale (2-task Cityscapes and 3-task NYUv2) and large-scale (16-task Taskonomy) datasets show that AutoTaskFormer outperforms state-of-the-art handcrafted vision transformers in multi-task learning. The entire code and models will be open-sourced.

Abstract: Recent advances in implicit neural representation have demonstrated the ability to recover detailed geometry and material from multi-view images. However, the use of simplified lighting models such as environment maps to represent non-distant illumination, or using a network to fit indirect light modeling without a solid basis, can lead to an undesirable decomposition between lighting and material. To address this, we propose a fully differentiable framework named neural ambient illumination (NeAI) that uses Neural Radiance Fields (NeRF) as a lighting model to handle complex lighting in a physically based way. Together with integral lobe encoding for roughness-adaptive specular lobe and leveraging the pre-convoluted background for accurate decomposition, the proposed method represents a significant step towards integrating physically based rendering into the NeRF representation. The experiments demonstrate the superior performance of novel-view rendering compared to previous works, and the capability to re-render objects under arbitrary NeRF-style environments opens up exciting possibilities for bridging the gap between virtual and real-world scenes. The project and supplementary materials are available at https://yiyuzhuang.github.io/NeAI/.

Abstract: In recent years, tremendous efforts have been made on document image rectification, but existing advanced algorithms are limited to processing restricted document images, i.e., the input images must incorporate a complete document. Once the captured image merely involves a local text region, its rectification quality is degraded and unsatisfactory. Our previously proposed DocTr, a transformer-assisted network for document image rectification, also suffers from this limitation. In this work, we present DocTr++, a novel unified framework for document image rectification, without any restrictions on the input distorted images. Our major technical improvements can be concluded in three aspects. Firstly, we upgrade the original architecture by adopting a hierarchical encoder-decoder structure for multi-scale representation extraction and parsing. Secondly, we reformulate the pixel-wise mapping relationship between the unrestricted distorted document images and the distortion-free counterparts. The obtained data is used to train our DocTr++ for unrestricted document image rectification. Thirdly, we contribute a real-world test set and metrics applicable for evaluating the rectification quality. To our best knowledge, this is the first learning-based method for the rectification of unrestricted document images. Extensive experiments are conducted, and the results demonstrate the effectiveness and superiority of our method. We hope our DocTr++ will serve as a strong baseline for generic document image rectification, prompting the further advancement and application of learning-based algorithms. The source code and the proposed dataset are publicly available at https://github.com/fh2019ustc/DocTr-Plus.

Abstract: 3D Skeleton-based human action recognition has attracted increasing attention in recent years. Most of the existing work focuses on supervised learning which requires a large number of labeled action sequences that are often expensive and time-consuming to annotate. In this paper, we address self-supervised 3D action representation learning for skeleton-based action recognition. We investigate self-supervised representation learning and design a novel skeleton cloud colorization technique that is capable of learning spatial and temporal skeleton representations from unlabeled skeleton sequence data. We represent a skeleton action sequence as a 3D skeleton cloud and colorize each point in the cloud according to its temporal and spatial orders in the original (unannotated) skeleton sequence. Leveraging the colorized skeleton point cloud, we design an auto-encoder framework that can learn spatial-temporal features from the artificial color labels of skeleton joints effectively. Specifically, we design a two-steam pretraining network that leverages fine-grained and coarse-grained colorization to learn multi-scale spatial-temporal features.In addition, we design a Masked Skeleton Cloud Repainting task that can pretrain the designed auto-encoder framework to learn informative representations. We evaluate our skeleton cloud colorization approach with linear classifiers trained under different configurations, including unsupervised, semi-supervised, fully-supervised, and transfer learning settings. Extensive experiments on NTU RGB+D, NTU RGB+D 120, PKU-MMD, NW-UCLA, and UWA3D datasets show that the proposed method outperforms existing unsupervised and semi-supervised 3D action recognition methods by large margins and achieves competitive performance in supervised 3D action recognition as well.

Abstract: The task of Group Activity Recognition (GAR) aims to predict the activity category of the group by learning the actor spatial-temporal interaction relation in the group. Therefore, an effective actor relation learning method is crucial for the GAR task. The previous works mainly learn the interaction relation by the well-designed GCNs or Transformers. For example, to infer the actor interaction relation, GCNs need a learnable adjacency, and Transformers need to calculate the self-attention. Although the above methods can model the interaction relation effectively, they also increase the complexity of the model (the number of parameters and computations). In this paper, we design a novel MLP-based method for Actor Interaction Relation learning (MLP-AIR) in GAR. Compared with GCNs and Transformers, our method has a competitive but conceptually and technically simple alternative, significantly reducing the complexity. Specifically, MLP-AIR includes three sub-modules: MLP-based Spatial relation modeling module (MLP-S), MLP-based Temporal relation modeling module (MLP-T), and MLP-based Relation refining module (MLP-R). MLP-S is used to model the spatial relation between different actors in each frame. MLP-T is used to model the temporal relation between different frames for each actor. MLP-R is used further to refine the relation between different dimensions of relation features to improve the feature's expression ability. To evaluate the MLP-AIR, we conduct extensive experiments on two widely used benchmarks, including the Volleyball and Collective Activity datasets. Experimental results demonstrate that MLP-AIR can get competitive results but with low complexity.

Abstract: Do video-text transformers learn to model temporal relationships across frames? Despite their immense capacity and the abundance of multimodal training data, recent work has revealed the strong tendency of video-text models towards frame-based spatial representations, while temporal reasoning remains largely unsolved. In this work, we identify several key challenges in temporal learning of video-text transformers: the spatiotemporal trade-off from limited network size; the curse of dimensionality for multi-frame modeling; and the diminishing returns of semantic information by extending clip length. Guided by these findings, we propose SViTT, a sparse video-text architecture that performs multi-frame reasoning with significantly lower cost than naive transformers with dense attention. Analogous to graph-based networks, SViTT employs two forms of sparsity: edge sparsity that limits the query-key communications between tokens in self-attention, and node sparsity that discards uninformative visual tokens. Trained with a curriculum which increases model sparsity with the clip length, SViTT outperforms dense transformer baselines on multiple video-text retrieval and question answering benchmarks, with a fraction of computational cost. Project page: http://svcl.ucsd.edu/projects/svitt.

Abstract: Latent Diffusion Models (LDMs) enable high-quality image synthesis while avoiding excessive compute demands by training a diffusion model in a compressed lower-dimensional latent space. Here, we apply the LDM paradigm to high-resolution video generation, a particularly resource-intensive task. We first pre-train an LDM on images only; then, we turn the image generator into a video generator by introducing a temporal dimension to the latent space diffusion model and fine-tuning on encoded image sequences, i.e., videos. Similarly, we temporally align diffusion model upsamplers, turning them into temporally consistent video super resolution models. We focus on two relevant real-world applications: Simulation of in-the-wild driving data and creative content creation with text-to-video modeling. In particular, we validate our Video LDM on real driving videos of resolution 512 x 1024, achieving state-of-the-art performance. Furthermore, our approach can easily leverage off-the-shelf pre-trained image LDMs, as we only need to train a temporal alignment model in that case. Doing so, we turn the publicly available, state-of-the-art text-to-image LDM Stable Diffusion into an efficient and expressive text-to-video model with resolution up to 1280 x 2048. We show that the temporal layers trained in this way generalize to different fine-tuned text-to-image LDMs. Utilizing this property, we show the first results for personalized text-to-video generation, opening exciting directions for future content creation. Project page: https://research.nvidia.com/labs/toronto-ai/VideoLDM/

Abstract: In recent years, video semantic segmentation has made great progress with advanced deep neural networks. However, there still exist two main challenges \ie, information inconsistency and computation cost. To deal with the two difficulties, we propose a novel motion-state alignment framework for video semantic segmentation to keep both motion and state consistency. In the framework, we first construct a motion alignment branch armed with an efficient decoupled transformer to capture dynamic semantics, guaranteeing region-level temporal consistency. Then, a state alignment branch composed of a stage transformer is designed to enrich feature spaces for the current frame to extract static semantics and achieve pixel-level state consistency. Next, by a semantic assignment mechanism, the region descriptor of each semantic category is gained from dynamic semantics and linked with pixel descriptors from static semantics. Benefiting from the alignment of these two kinds of effective information, the proposed method picks up dynamic and static semantics in a targeted way, so that video semantic regions are consistently segmented to obtain precise locations with low computational complexity. Extensive experiments on Cityscapes and CamVid datasets show that the proposed approach outperforms state-of-the-art methods and validates the effectiveness of the motion-state alignment framework.

Abstract: Data augmentation has been established as an efficacious approach to supplement useful information for low-resource datasets. Traditional augmentation techniques such as noise injection and image transformations have been widely used. In addition, generative data augmentation (GDA) has been shown to produce more diverse and flexible data. While generative adversarial networks (GANs) have been frequently used for GDA, they lack diversity and controllability compared to text-to-image diffusion models. In this paper, we propose TTIDA (Text-to-Text-to-Image Data Augmentation) to leverage the capabilities of large-scale pre-trained Text-to-Text (T2T) and Text-to-Image (T2I) generative models for data augmentation. By conditioning the T2I model on detailed descriptions produced by T2T models, we are able to generate photo-realistic labeled images in a flexible and controllable manner. Experiments on in-domain classification, cross-domain classification, and image captioning tasks show consistent improvements over other data augmentation baselines. Analytical studies in varied settings, including few-shot, long-tail, and adversarial, further reinforce the effectiveness of TTIDA in enhancing performance and increasing robustness.

Abstract: Recently, instance segmentation has made great progress with the rapid development of deep neural networks. However, there still exist two main challenges including discovering indistinguishable objects and modeling the relationship between instances. To deal with these difficulties, we propose a novel object mining framework for instance segmentation. In this framework, we first introduce the semantics perceiving subnetwork to capture pixels that may belong to an obvious instance from the bottom up. Then, we propose an object excavating mechanism to discover indistinguishable objects. In the mechanism, preliminary perceived semantics are regarded as original instances with classifications and locations, and then indistinguishable objects around these original instances are mined, which ensures that hard objects are fully excavated. Next, an instance purifying strategy is put forward to model the relationship between instances, which pulls the similar instances close and pushes away different instances to keep intra-instance similarity and inter-instance discrimination. In this manner, the same objects are combined as the one instance and different objects are distinguished as independent instances. Extensive experiments on the COCO dataset show that the proposed approach outperforms state-of-the-art methods, which validates the effectiveness of the proposed object mining framework.

Abstract: It is seen that there is enormous potential to leverage powerful deep learning methods in the emerging field of urban digital twins. It is particularly in the area of intelligent road inspection where there is currently limited research and data available. To facilitate progress in this field, we have developed a well-labeled road pothole dataset named Urban Digital Twins Intelligent Road Inspection (UDTIRI) dataset. We hope this dataset will enable the use of powerful deep learning methods in urban road inspection, providing algorithms with a more comprehensive understanding of the scene and maximizing their potential. Our dataset comprises 1000 images of potholes, captured in various scenarios with different lighting and humidity conditions. Our intention is to employ this dataset for object detection, semantic segmentation, and instance segmentation tasks. Our team has devoted significant effort to conducting a detailed statistical analysis, and benchmarking a selection of representative algorithms from recent years. We also provide a multi-task platform for researchers to fully exploit the performance of various algorithms with the support of UDTIRI dataset.

Abstract: Stereo video retargeting aims to resize an image to a desired aspect ratio. The quality of retargeted videos can be significantly impacted by the stereo videos spatial, temporal, and disparity coherence, all of which can be impacted by the retargeting process. Due to the lack of a publicly accessible annotated dataset, there is little research on deep learning-based methods for stereo video retargeting. This paper proposes an unsupervised deep learning-based stereo video retargeting network. Our model first detects the salient objects and shifts and warps all objects such that it minimizes the distortion of the salient parts of the stereo frames. We use 1D convolution for shifting the salient objects and design a stereo video Transformer to assist the retargeting process. To train the network, we use the parallax attention mechanism to fuse the left and right views and feed the retargeted frames to a reconstruction module that reverses the retargeted frames to the input frames. Therefore, the network is trained in an unsupervised manner. Extensive qualitative and quantitative experiments and ablation studies on KITTI stereo 2012 and 2015 datasets demonstrate the efficiency of the proposed method over the existing state-of-the-art methods. The code is available at https://github.com/z65451/SVR/.

Abstract: Existing person image generative models can do either image generation or pose transfer but not both. We propose a unified diffusion model, UPGPT to provide a universal solution to perform all the person image tasks - generative, pose transfer, and editing. With fine-grained multimodality and disentanglement capabilities, our approach offers fine-grained control over the generation and the editing process of images using a combination of pose, text, and image, all without needing a semantic segmentation mask which can be challenging to obtain or edit. We also pioneer the parameterized body SMPL model in pose-guided person image generation to demonstrate new capability - simultaneous pose and camera view interpolation while maintaining a person's appearance. Results on the benchmark DeepFashion dataset show that UPGPT is the new state-of-the-art while simultaneously pioneering new capabilities of edit and pose transfer in human image generation.

Abstract: Detecting arbitrarily oriented tiny objects poses intense challenges to existing detectors, especially for label assignment. Despite the exploration of adaptive label assignment in recent oriented object detectors, the extreme geometry shape and limited feature of oriented tiny objects still induce severe mismatch and imbalance issues. Specifically, the position prior, positive sample feature, and instance are mismatched, and the learning of extreme-shaped objects is biased and unbalanced due to little proper feature supervision. To tackle these issues, we propose a dynamic prior along with the coarse-to-fine assigner, dubbed DCFL. For one thing, we model the prior, label assignment, and object representation all in a dynamic manner to alleviate the mismatch issue. For another, we leverage the coarse prior matching and finer posterior constraint to dynamically assign labels, providing appropriate and relatively balanced supervision for diverse instances. Extensive experiments on six datasets show substantial improvements to the baseline. Notably, we obtain the state-of-the-art performance for one-stage detectors on the DOTA-v1.5, DOTA-v2.0, and DIOR-R datasets under single-scale training and testing. Codes are available at https://github.com/Chasel-Tsui/mmrotate-dcfl.

Abstract: Many existing studies on knowledge distillation have focused on methods in which a student model mimics a teacher model well. Simply imitating the teacher's knowledge, however, is not sufficient for the student to surpass that of the teacher. We explore a method to harness the knowledge of other students to complement the knowledge of the teacher. We propose deep collective knowledge distillation for model compression, called DCKD, which is a method for training student models with rich information to acquire knowledge from not only their teacher model but also other student models. The knowledge collected from several student models consists of a wealth of information about the correlation between classes. Our DCKD considers how to increase the correlation knowledge of classes during training. Our novel method enables us to create better performing student models for collecting knowledge. This simple yet powerful method achieves state-of-the-art performances in many experiments. For example, for ImageNet, ResNet18 trained with DCKD achieves 72.27\%, which outperforms the pretrained ResNet18 by 2.52\%. For CIFAR-100, the student model of ShuffleNetV1 with DCKD achieves 6.55\% higher top-1 accuracy than the pretrained ShuffleNetV1.

Abstract: Self-supervised monocular depth estimation approaches suffer not only from scale ambiguity but also infer temporally inconsistent depth maps w.r.t. scale. While disambiguating scale during training is not possible without some kind of ground truth supervision, having scale consistent depth predictions would make it possible to calculate scale once during inference as a post-processing step and use it over-time. With this as a goal, a set of temporal consistency losses that minimize pose inconsistencies over time are introduced. Evaluations show that introducing these constraints not only reduces depth inconsistencies but also improves the baseline performance of depth and ego-motion prediction.

Abstract: This paper proposes a quantum computing-based algorithm to solve the single image super-resolution (SISR) problem. One of the well-known classical approaches for SISR relies on the well-established patch-wise sparse modeling of the problem. Yet, this field's current state of affairs is that deep neural networks (DNNs) have demonstrated far superior results than traditional approaches. Nevertheless, quantum computing is expected to become increasingly prominent for machine learning problems soon. As a result, in this work, we take the privilege to perform an early exploration of applying a quantum computing algorithm to this important image enhancement problem, i.e., SISR. Among the two paradigms of quantum computing, namely universal gate quantum computing and adiabatic quantum computing (AQC), the latter has been successfully applied to practical computer vision problems, in which quantum parallelism has been exploited to solve combinatorial optimization efficiently. This work demonstrates formulating quantum SISR as a sparse coding optimization problem, which is solved using quantum annealers accessed via the D-Wave Leap platform. The proposed AQC-based algorithm is demonstrated to achieve improved speed-up over a classical analog while maintaining comparable SISR accuracy.

Abstract: Learning-based point cloud registration methods can handle clean point clouds well, while it is still challenging to generalize to noisy and partial point clouds. To this end, we propose a novel framework for noisy and partial point cloud registration. By introducing a neural implicit function representation, we replace the problem of rigid registration between point clouds with a registration problem between the point cloud and the neural implicit function. We then alternately optimize the implicit function representation and the registration between the implicit function and point cloud. In this way, point cloud registration can be performed in a coarse-to-fine manner. Since our method avoids computing point correspondences, it is robust to the noise and incompleteness of point clouds. Compared with the registration methods based on global features, our method can deal with surfaces with large density variations and achieve higher registration accuracy. Experimental results and comparisons demonstrate the effectiveness of the proposed framework.

Abstract: Textbooks are the primary vehicle for delivering quality education to students. It has been shown that explanatory or illustrative visuals play a key role in the retention, comprehension and the general transfer of knowledge. However, many textbooks, especially in the developing world, are low quality and lack interesting visuals to support student learning. In this paper, we investigate the effectiveness of vision-language models to automatically enhance textbooks with images from the web. Specifically, we collect a dataset of e-textbooks from one of the largest free online publishers in the world. We rigorously analyse the dataset, and use the resulting analysis to motivate a task that involves retrieving and appropriately assigning web images to textbooks, which we frame as a novel optimization problem. Through a crowd-sourced evaluation, we verify that (1) while the original textbook images are rated higher, automatically assigned ones are not far behind, and (2) the choice of the optimization problem matters. We release the dataset of textbooks with an associated image bank to spur further research in this area.

Abstract: Facial expression editing has attracted increasing attention with the advance of deep neural networks in recent years. However, most existing methods suffer from compromised editing fidelity and limited usability as they either ignore pose variations (unrealistic editing) or require paired training data (not easy to collect) for pose controls. This paper presents POCE, an innovative pose-controllable expression editing network that can generate realistic facial expressions and head poses simultaneously with just unpaired training images. POCE achieves the more accessible and realistic pose-controllable expression editing by mapping face images into UV space, where facial expressions and head poses can be disentangled and edited separately. POCE has two novel designs. The first is self-supervised UV completion that allows to complete UV maps sampled under different head poses, which often suffer from self-occlusions and missing facial texture. The second is weakly-supervised UV editing that allows to generate new facial expressions with minimal modification of facial identity, where the synthesized expression could be controlled by either an expression label or directly transplanted from a reference UV map via feature transfer. Extensive experiments show that POCE can learn from unpaired face images effectively, and the learned model can generate realistic and high-fidelity facial expressions under various new poses.

Abstract: Audio-driven talking face generation, which aims to synthesize talking faces with realistic facial animations (including accurate lip movements, vivid facial expression details and natural head poses) corresponding to the audio, has achieved rapid progress in recent years. However, most existing work focuses on generating lip movements only without handling the closely correlated facial expressions, which degrades the realism of the generated faces greatly. This paper presents DIRFA, a novel method that can generate talking faces with diverse yet realistic facial animations from the same driving audio. To accommodate fair variation of plausible facial animations for the same audio, we design a transformer-based probabilistic mapping network that can model the variational facial animation distribution conditioned upon the input audio and autoregressively convert the audio signals into a facial animation sequence. In addition, we introduce a temporally-biased mask into the mapping network, which allows to model the temporal dependency of facial animations and produce temporally smooth facial animation sequence. With the generated facial animation sequence and a source image, photo-realistic talking faces can be synthesized with a generic generation network. Extensive experiments show that DIRFA can generate talking faces with realistic facial animations effectively.

Abstract: Virtual try-on is a promising computer vision topic with a high commercial value wherein a new garment is visually worn on a person with a photo-realistic effect. Previous studies conduct their shape and content inference at one stage, employing a single-scale warping mechanism and a relatively unsophisticated content inference mechanism. These approaches have led to suboptimal results in terms of garment warping and skin reservation under challenging try-on scenarios. To address these limitations, we propose a novel virtual try-on method via progressive inference paradigm (PGVTON) that leverages a top-down inference pipeline and a general garment try-on strategy. Specifically, we propose a robust try-on parsing inference method by disentangling semantic categories and introducing consistency. Exploiting the try-on parsing as the shape guidance, we implement the garment try-on via warping-mapping-composition. To facilitate adaptation to a wide range of try-on scenarios, we adopt a covering more and selecting one warping strategy and explicitly distinguish tasks based on alignment. Additionally, we regulate StyleGAN2 to implement re-naked skin inpainting, conditioned on the target skin shape and spatial-agnostic skin features. Experiments demonstrate that our method has state-of-the-art performance under two challenging scenarios. The code will be available at https://github.com/NerdFNY/PGVTON.

Abstract: Data-driven cell tracking and segmentation methods in biomedical imaging require diverse and information-rich training data. In cases where the number of training samples is limited, synthetic computer-generated data sets can be used to improve these methods. This requires the synthesis of cell shapes as well as corresponding microscopy images using generative models. To synthesize realistic living cell shapes, the shape representation used by the generative model should be able to accurately represent fine details and changes in topology, which are common in cells. These requirements are not met by 3D voxel masks, which are restricted in resolution, and polygon meshes, which do not easily model processes like cell growth and mitosis. In this work, we propose to represent living cell shapes as level sets of signed distance functions (SDFs) which are estimated by neural networks. We optimize a fully-connected neural network to provide an implicit representation of the SDF value at any point in a 3D+time domain, conditioned on a learned latent code that is disentangled from the rotation of the cell shape. We demonstrate the effectiveness of this approach on cells that exhibit rapid deformations (Platynereis dumerilii), cells that grow and divide (C. elegans), and cells that have growing and branching filopodial protrusions (A549 human lung carcinoma cells). A quantitative evaluation using shape features, Hausdorff distance, and Dice similarity coefficients of real and synthetic cell shapes shows that our model can generate topologically plausible complex cell shapes in 3D+time with high similarity to real living cell shapes. Finally, we show how microscopy images of living cells that correspond to our generated cell shapes can be synthesized using an image-to-image model.

Abstract: Semantic segmentation of point clouds usually requires exhausting efforts of human annotations, hence it attracts wide attention to the challenging topic of learning from unlabeled or weaker forms of annotations. In this paper, we take the first attempt for fully unsupervised semantic segmentation of point clouds, which aims to delineate semantically meaningful objects without any form of annotations. Previous works of unsupervised pipeline on 2D images fails in this task of point clouds, due to: 1) Clustering Ambiguity caused by limited magnitude of data and imbalanced class distribution; 2) Irregularity Ambiguity caused by the irregular sparsity of point cloud. Therefore, we propose a novel framework, PointDC, which is comprised of two steps that handle the aforementioned problems respectively: Cross-Modal Distillation (CMD) and Super-Voxel Clustering (SVC). In the first stage of CMD, multi-view visual features are back-projected to the 3D space and aggregated to a unified point feature to distill the training of the point representation. In the second stage of SVC, the point features are aggregated to super-voxels and then fed to the iterative clustering process for excavating semantic classes. PointDC yields a significant improvement over the prior state-of-the-art unsupervised methods, on both the ScanNet-v2 (+18.4 mIoU) and S3DIS (+11.5 mIoU) semantic segmentation benchmarks.

Abstract: Online reconstructing and rendering of large-scale indoor scenes is a long-standing challenge. SLAM-based methods can reconstruct 3D scene geometry progressively in real time but can not render photorealistic results. While NeRF-based methods produce promising novel view synthesis results, their long offline optimization time and lack of geometric constraints pose challenges to efficiently handling online input. Inspired by the complementary advantages of classical 3D reconstruction and NeRF, we thus investigate marrying explicit geometric representation with NeRF rendering to achieve efficient online reconstruction and high-quality rendering. We introduce SurfelNeRF, a variant of neural radiance field which employs a flexible and scalable neural surfel representation to store geometric attributes and extracted appearance features from input images. We further extend the conventional surfel-based fusion scheme to progressively integrate incoming input frames into the reconstructed global neural scene representation. In addition, we propose a highly-efficient differentiable rasterization scheme for rendering neural surfel radiance fields, which helps SurfelNeRF achieve $10\times$ speedups in training and inference time, respectively. Experimental results show that our method achieves the state-of-the-art 23.82 PSNR and 29.58 PSNR on ScanNet in feedforward inference and per-scene optimization settings, respectively.

Abstract: This paper presents AutoPatch, the first application of neural architecture search to the complex task of segmenting visual anomalies. Measurement of anomaly segmentation quality is challenging due to imbalanced anomaly pixels, varying region areas, and various types of anomalies. First, the weighted average precision (wAP) metric is proposed as an alternative to AUROC and AUPRO, which does not need to be limited to a specific maximum FPR. Second, a novel neural architecture search method is proposed, which enables efficient segmentation of visual anomalies without any training. By leveraging a pre-trained supernet, a black-box optimization algorithm can directly minimize FLOPS and maximize wAP on a small validation set of anomalous examples. Finally, compelling results on the widely studied MVTec [3] dataset are presented, demonstrating that AutoPatch outperforms the current state-of-the-art method PatchCore [12] with more than 18x fewer FLOPS, using only one example per anomaly type. These results highlight the potential of automated machine learning to optimize throughput in industrial quality control. The code for AutoPatch is available at: https://github.com/tommiekerssies/AutoPatch

Abstract: This paper presents a novel visual-LiDAR odometry and mapping method with low-drift characteristics. The proposed method is based on two popular approaches, ORB-SLAM and A-LOAM, with monocular scale correction and visual-assisted LiDAR motion compensation modifications. The scale corrector calculates the proportion between the depth of image keypoints recovered by triangulation and that provided by LiDAR, using an outlier rejection process for accuracy improvement. Concerning LiDAR motion compensation, the visual odometry approach gives the initial guesses of LiDAR motions for better performance. This methodology is not only applicable to high-resolution LiDAR but can also adapt to low-resolution LiDAR. To evaluate the proposed SLAM system's robustness and accuracy, we conducted experiments on the KITTI Odometry and S3E datasets. Experimental results illustrate that our method significantly outperforms standalone ORB-SLAM2 and A-LOAM. Furthermore, regarding the accuracy of visual odometry with scale correction, our method performs similarly to the stereo-mode ORB-SLAM2.

Abstract: As the use of deep neural networks continues to grow, understanding their behaviour has become more crucial than ever. Post-hoc explainability methods are a potential solution, but their reliability is being called into question. Our research investigates the response of post-hoc visual explanations to naturally occurring transformations, often referred to as augmentations. We anticipate explanations to be invariant under certain transformations, such as changes to the colour map while responding in an equivariant manner to transformations like translation, object scaling, and rotation. We have found remarkable differences in robustness depending on the type of transformation, with some explainability methods (such as LRP composites and Guided Backprop) being more stable than others. We also explore the role of training with data augmentation. We provide evidence that explanations are typically less robust to augmentation than classification performance, regardless of whether data augmentation is used in training or not.

Abstract: Data quality is critical for multimedia tasks, while various types of systematic flaws are found in image benchmark datasets, as discussed in recent work. In particular, the existence of the semantic gap problem leads to a many-to-many mapping between the information extracted from an image and its linguistic description. This unavoidable bias further leads to poor performance on current computer vision tasks. To address this issue, we introduce a Knowledge Representation (KR)-based methodology to provide guidelines driving the labeling process, thereby indirectly introducing intended semantics in ML models. Specifically, an iterative refinement-based annotation method is proposed to optimize data labeling by organizing objects in a classification hierarchy according to their visual properties, ensuring that they are aligned with their linguistic descriptions. Preliminary results verify the effectiveness of the proposed method.

Abstract: Multi-frame depth estimation generally achieves high accuracy relying on the multi-view geometric consistency. When applied in dynamic scenes, e.g., autonomous driving, this consistency is usually violated in the dynamic areas, leading to corrupted estimations. Many multi-frame methods handle dynamic areas by identifying them with explicit masks and compensating the multi-view cues with monocular cues represented as local monocular depth or features. The improvements are limited due to the uncontrolled quality of the masks and the underutilized benefits of the fusion of the two types of cues. In this paper, we propose a novel method to learn to fuse the multi-view and monocular cues encoded as volumes without needing the heuristically crafted masks. As unveiled in our analyses, the multi-view cues capture more accurate geometric information in static areas, and the monocular cues capture more useful contexts in dynamic areas. To let the geometric perception learned from multi-view cues in static areas propagate to the monocular representation in dynamic areas and let monocular cues enhance the representation of multi-view cost volume, we propose a cross-cue fusion (CCF) module, which includes the cross-cue attention (CCA) to encode the spatially non-local relative intra-relations from each source to enhance the representation of the other. Experiments on real-world datasets prove the significant effectiveness and generalization ability of the proposed method.

Abstract: We focus on enabling damage and tampering detection in logistics and tackle the problem of 3D shape reconstruction of potentially damaged parcels. As input we utilize single RGB images, which corresponds to use-cases where only simple handheld devices are available, e.g. for postmen during delivery or clients on delivery. We present a novel synthetic dataset, named Parcel3D, that is based on the Google Scanned Objects (GSO) dataset and consists of more than 13,000 images of parcels with full 3D annotations. The dataset contains intact, i.e. cuboid-shaped, parcels and damaged parcels, which were generated in simulations. We work towards detecting mishandling of parcels by presenting a novel architecture called CubeRefine R-CNN, which combines estimating a 3D bounding box with an iterative mesh refinement. We benchmark our approach on Parcel3D and an existing dataset of cuboid-shaped parcels in real-world scenarios. Our results show, that while training on Parcel3D enables transfer to the real world, enabling reliable deployment in real-world scenarios is still challenging. CubeRefine R-CNN yields competitive performance in terms of Mesh AP and is the only model that directly enables deformation assessment by 3D mesh comparison and tampering detection by comparing viewpoint invariant parcel side surface representations. Dataset and code are available at https://a-nau.github.io/parcel3d.

Abstract: Sketching out Graphical User Interface (GUI) layout is part of the pipeline of designing a GUI and a crucial task for the success of a software application. Arranging all components inside a GUI layout manually is a time-consuming task. In order to assist designers, we developed a method named GUILGET to automatically generate GUI layouts from positional constraints represented as GUI arrangement graphs (GUI-AGs). The goal is to support the initial step of GUI design by producing realistic and diverse GUI layouts. The existing image layout generation techniques often cannot incorporate GUI design constraints. Thus, GUILGET needs to adapt existing techniques to generate GUI layouts that obey to constraints specific to GUI designs. GUILGET is based on transformers in order to capture the semantic in relationships between elements from GUI-AG. Moreover, the model learns constraints through the minimization of losses responsible for placing each component inside its parent layout, for not letting components overlap if they are inside the same parent, and for component alignment. Our experiments, which are conducted on the CLAY dataset, reveal that our model has the best understanding of relationships from GUI-AG and has the best performances in most of evaluation metrics. Therefore, our work contributes to improved GUI layout generation by proposing a novel method that effectively accounts for the constraints on GUI elements and paves the road for a more efficient GUI design pipeline.

Abstract: Generative adversarial networks (GANs) are successfully used for image synthesis but are known to face instability during training. In contrast, probabilistic diffusion models (DMs) are stable and generate high-quality images, at the cost of an expensive sampling procedure. In this paper, we introduce a simple method to allow GANs to stably converge to their theoretical optimum, while bringing in the denoising machinery from DMs. These models are combined into a simpler model (ATME) that only requires a forward pass during inference, making predictions cheaper and more accurate than DMs and popular GANs. ATME breaks an information asymmetry existing in most GAN models in which the discriminator has spatial knowledge of where the generator is failing. To restore the information symmetry, the generator is endowed with knowledge of the entropic state of the discriminator, which is leveraged to allow the adversarial game to converge towards equilibrium. We demonstrate the power of our method in several image-to-image translation tasks, showing superior performance than state-of-the-art methods at a lesser cost. Code is available at https://github.com/DLR-MI/atme

Abstract: Currently intelligent diagnosis systems lack the ability of continually learning to diagnose new diseases once deployed, under the condition of preserving old disease knowledge. In particular, updating an intelligent diagnosis system with training data of new diseases would cause catastrophic forgetting of old disease knowledge. To address the catastrophic forgetting issue, a novel adapter-based strategy is proposed to help effectively learn a set of new diseases at each round (or task) of continual learning, without changing the shared feature extractor. The learnable lightweight task-specific adapter(s) can be flexibly designed (e.g., two convolutional layers) and then added to the pretrained and fixed feature extractor. Together with a specially designed task-specific head which absorbs all previously learned old diseases as a single 'out-of-distribution' category, task-specific adapter(s) can help the pretrained feature extractor more effectively extract discriminative features between diseases. In addition, a simple yet effective fine-tuning is applied to collaboratively fine-tune multiple task-specific heads such that outputs from different heads are comparable and consequently the appropriate classifier head can be more accurately selected during model inference. Extensive empirical evaluations on three image datasets demonstrate the superior performance of the proposed method in continual learning of new diseases. The source code will be released publicly.

Abstract: Thanks to the advantages of the friendly annotations and the satisfactory performance, Weakly-Supervised Semantic Segmentation (WSSS) approaches have been extensively studied. Recently, the single-stage WSSS was awakened to alleviate problems of the expensive computational costs and the complicated training procedures in multi-stage WSSS. However, results of such an immature model suffer from problems of \emph{background incompleteness} and \emph{object incompleteness}. We empirically find that they are caused by the insufficiency of the global object context and the lack of the local regional contents, respectively. Under these observations, we propose a single-stage WSSS model with only the image-level class label supervisions, termed as \textbf{W}eakly-\textbf{S}upervised \textbf{F}eature \textbf{C}oupling \textbf{N}etwork (\textbf{WS-FCN}), which can capture the multi-scale context formed from the adjacent feature grids, and encode the fine-grained spatial information from the low-level features into the high-level ones. Specifically, a flexible context aggregation module is proposed to capture the global object context in different granular spaces. Besides, a semantically consistent feature fusion module is proposed in a bottom-up parameter-learnable fashion to aggregate the fine-grained local contents. Based on these two modules, \textbf{WS-FCN} lies in a self-supervised end-to-end training fashion. Extensive experimental results on the challenging PASCAL VOC 2012 and MS COCO 2014 demonstrate the effectiveness and efficiency of \textbf{WS-FCN}, which can achieve state-of-the-art results by $65.02\%$ and $64.22\%$ mIoU on PASCAL VOC 2012 \emph{val} set and \emph{test} set, $34.12\%$ mIoU on MS COCO 2014 \emph{val} set, respectively. The code and weight have been released at:~\href{https://github.com/ChunyanWang1/ws-fcn}{WS-FCN}.

Abstract: Endobronchial intervention is increasingly used as a minimally invasive means for the treatment of pulmonary diseases. In order to reduce the difficulty of manipulation in complex airway networks, robust lumen detection is essential for intraoperative guidance. However, these methods are sensitive to visual artifacts which are inevitable during the surgery. In this work, a cross domain feature interaction (CDFI) network is proposed to extract the structural features of lumens, as well as to provide artifact cues to characterize the visual features. To effectively extract the structural and artifact features, the Quadruple Feature Constraints (QFC) module is designed to constrain the intrinsic connections of samples with various imaging-quality. Furthermore, we design a Guided Feature Fusion (GFF) module to supervise the model for adaptive feature fusion based on different types of artifacts. Results show that the features extracted by the proposed method can preserve the structural information of lumen in the presence of large visual variations, bringing much-improved lumen detection accuracy.

Abstract: Scene flow is an important problem as it provides low-level motion cues for many downstream tasks. State-of-the-art learning methods are usually fast and can achieve impressive performance on in-domain data, but usually fail to generalize to out-of-the-distribution (OOD) data or handle dense point clouds. In this paper, we focus on a runtime optimization-based neural scene flow pipeline. In (a) one can see its application in the densification of lidar. However, in (c) one sees that the major drawback is the extensive computation time. We identify that the common speedup strategy in network architectures for coordinate networks has little effect on scene flow acceleration [see green (b)] unlike image reconstruction [see pink (b)]. With the dominant computational burden stemming instead from the Chamfer loss function, we propose to use a distance transform-based loss function to accelerate [see purple (b)], which achieves up to 30x speedup and on-par estimation performance compared to NSFP [see (c)]. When tested on 8k points, it is as efficient [see (c)] as leading learning methods, achieving real-time performance.

Abstract: Real-scanned point clouds are often incomplete due to viewpoint, occlusion, and noise, which hampers 3D geometric modeling and perception. Existing point cloud completion methods tend to generate global shape skeletons and hence lack fine local details. Furthermore, they mostly learn a deterministic partial-to-complete mapping, but overlook structural relations in man-made objects. To tackle these challenges, this paper proposes a variational framework, Variational Relational point Completion Network (VRCNet) with two appealing properties: 1) Probabilistic Modeling. In particular, we propose a dual-path architecture to enable principled probabilistic modeling across partial and complete clouds. One path consumes complete point clouds for reconstruction by learning a point VAE. The other path generates complete shapes for partial point clouds, whose embedded distribution is guided by distribution obtained from the reconstruction path during training. 2) Relational Enhancement. Specifically, we carefully design point self-attention kernel and point selective kernel module to exploit relational point features, which refines local shape details conditioned on the coarse completion. In addition, we contribute multi-view partial point cloud datasets (MVP and MVP-40 dataset) containing over 200,000 high-quality scans, which render partial 3D shapes from 26 uniformly distributed camera poses for each 3D CAD model. Extensive experiments demonstrate that VRCNet outperforms state-of-the-art methods on all standard point cloud completion benchmarks. Notably, VRCNet shows great generalizability and robustness on real-world point cloud scans. Moreover, we can achieve robust 3D classification for partial point clouds with the help of VRCNet, which can highly increase classification accuracy.

Abstract: The emergence of large models, also known as foundation models, has brought significant advancements to AI research. One such model is Segment Anything (SAM), which is designed for image segmentation tasks. However, as with other foundation models, our experimental findings suggest that SAM may fail or perform poorly in certain segmentation tasks, such as shadow detection and camouflaged object detection (concealed object detection). This study first paves the way for applying the large pre-trained image segmentation model SAM to these downstream tasks, even in situations where SAM performs poorly. Rather than fine-tuning the SAM network, we propose \textbf{SAM-Adapter}, which incorporates domain-specific information or visual prompts into the segmentation network by using simple yet effective adapters. Our extensive experiments show that SAM-Adapter can significantly elevate the performance of SAM in challenging tasks and we can even outperform task-specific network models and achieve state-of-the-art performance in the task we tested: camouflaged object detection and shadow detection. We believe our work opens up opportunities for utilizing SAM in downstream tasks, with potential applications in various fields, including medical image processing, agriculture, remote sensing, and more.

Abstract: Visual and linguistic concepts naturally organize themselves in a hierarchy, where a textual concept ``dog'' entails all images that contain dogs. Despite being intuitive, current large-scale vision and language models such as CLIP do not explicitly capture such hierarchy. We propose MERU, a contrastive model that yields hyperbolic representations of images and text. Hyperbolic spaces have suitable geometric properties to embed tree-like data, so MERU can better capture the underlying hierarchy in image-text data. Our results show that MERU learns a highly interpretable representation space while being competitive with CLIP's performance on multi-modal tasks like image classification and image-text retrieval.