ICCV 2023 论文列表

Recently, the RGB images and point clouds fusion methods have been proposed to jointly estimate 2D optical flow and 3D scene flow. However, as both conventional RGB cameras and LiDAR sensors adopt a frame-based data acquisition mechanism, their performance is limited by the fixed low sampling rates, especially in highly-dynamic scenes. By contrast, the event camera can asynchronously capture the intensity changes with a very high temporal resolution, providing complementary dynamic information of the observed scenes. In this paper, we incorporate RGB images, Point clouds and Events for joint optical flow and scene flow estimation with our proposed multi-stage multimodal fusion model, RPEFlow. First, we present an attention fusion module with a cross-attention mechanism to implicitly explore the internal cross-modal correlation for 2D and 3D branches, respectively. Second, we introduce a mutual information regularization term to explicitly model the complementary information of three modalities for effective multimodal feature learning. We also contribute a new synthetic dataset to advocate further research. Experiments on both synthetic and real datasets show that our model outperforms the existing state-of-the-art by a wide margin. Code and dataset is available at https://npucvr.github.io/RPEFlow.

Albeit being a prevalent architecture searching approach, differentiable architecture search (DARTS) is largely hindered by its substantial memory cost since the entire supernet resides in the memory. This is where the single-path DARTS comes in, which only chooses a single-path submodel at each step. While being memory-friendly, it also comes with low computational costs. Nonetheless, we discover a critical issue of single-path DARTS that has not been primarily noticed. Namely, it also suffers from severe performance collapse since too many parameter-free operations like skip connections are derived, just like DARTS does. In this paper, we propose a new algorithm called RObustifying Memory-Efficient NAS (ROME) to give a cure. First, we disentangle the topology search from the operation search to make searching and evaluation consistent. We then adopt Gumbel-Top2 reparameterization and gradient accumulation to robustify the unwieldy bi-level optimization. We verify ROME extensively across 15 benchmarks to demonstrate its effectiveness and robustness.

Spiking Neural Networks (SNNs) as one of the biology-inspired models have received much attention recently. It can significantly reduce energy consumption since they quantize the real-valued membrane potentials to 0/1 spikes to transmit information thus the multiplications of activations and weights can be replaced by additions when implemented on hardware. However, this quantization mechanism will inevitably introduce quantization error, thus causing catastrophic information loss. To address the quantization error problem, we propose a regularizing membrane potential loss (RMP-Loss) to adjust the distribution which is directly related to quantization error to a range close to the spikes. Our method is extremely simple to implement and straightforward to train an SNN. Furthermore, it is shown to consistently outperform previous state-of-the-art methods over different network architectures and datasets.

Robust estimation is a crucial and still challenging task, which involves estimating model parameters in noisy environments. Although conventional sampling consensus-based algorithms sample several times to achieve robustness, these algorithms cannot use data features and historical information effectively. In this paper, we propose RLSAC, a novel Reinforcement Learning enhanced SAmple Consensus framework for end-to-end robust estimation. RLSAC employs a graph neural network to utilize both data and memory features to guide exploring directions for sampling the next minimum set. The feedback of downstream tasks serves as the reward for unsupervised training. Therefore, RLSAC can avoid differentiating to learn the features and the feedback of downstream tasks for end-to-end robust estimation. In addition, RLSAC integrates a state transition module that encodes both data and memory features. Our experimental results demonstrate that RLSAC can learn from features to gradually explore a better hypothesis. Through analysis, it is apparent that RLSAC can be easily transferred to other sampling consensus-based robust estimation tasks. To the best of our knowledge, RLSAC is also the first method that uses reinforcement learning to sample consensus for end-to-end robust estimation. We release our codes at https://github.com/IRMVLab/RLSAC.

Relational Language-Image Pre-training (RLIP) aims to align vision representations with relational texts, thereby advancing the capability of relational reasoning in computer vision tasks. However, hindered by the slow convergence of RLIPv1 architecture and the limited availability of existing scene graph data, scaling RLIPv1 is challenging. In this paper, we propose RLIPv2, a fast converging model that enables the scaling of relational pre-training to large-scale pseudo-labelled scene graph data. To enable fast scaling, RLIPv2 introduces Asymmetric Language-Image Fusion (ALIF), a mechanism that facilitates earlier and deeper gated cross-modal fusion with sparsified language encoding layers. ALIF leads to comparable or better performance than RLIPv1 in a fraction of the time for pre-training and fine-tuning. To obtain scene graph data at scale, we extend object detection datasets with free-form relation labels by introducing a captioner (e.g., BLIP) and a designed Relation Tagger. The Relation Tagger assigns BLIP-generated relation texts to region pairs, thus enabling larger-scale relational pre-training. Through extensive experiments conducted on Human-Object Interaction Detection and Scene Graph Generation, RLIPv2 shows state-of-the-art performance on three benchmarks under fully-finetuning, few-shot and zero-shot settings. Notably, the largest RLIPv2 achieves 23.29mAP on HICO-DET without any fine-tuning, yields 32.22mAP with just 1% data and yields 45.09mAP with 100% data. Code and models are publicly available at https://github.com/JacobYuan7/RLIPv2.

GAN inversion is indispensable for applying the powerful editability of GAN to real images. However, existing methods invert video frames individually often leading to undesired inconsistent results over time. In this paper, we propose a unified recurrent framework, named Recurrent vIdeo GAN Inversion and eDiting (RIGID), to explicitly and simultaneously enforce temporally coherent GAN inversion and facial editing of real videos. Our approach models the temporal relations between current and previous frames from three aspects. To enable a faithful real video reconstruction, we first maximize the inversion fidelity and consistency by learning a temporal compensated latent code. Second, we observe incoherent noises lie in the high-frequency domain that can be disentangled from the latent space. Third, to remove the inconsistency after attribute manipulation, we propose an in-between frame composition constraint such that the arbitrary frame must be a direct composite of its neighboring frames. Our unified framework learns the inherent coherence between input frames in an end-to-end manner, and therefore it is agnostic to a specific attribute and can be applied to arbitrary editing of the same video without re-training. Extensive experiments demonstrate that RIGID outperforms state-of-the-art methods qualitatively and quantitatively in both inversion and editing tasks. The deliverables can be found in https://cnnlstm.github.io/RIGID.

Recently, neural implicit surfaces have become popular for multi-view reconstruction. To facilitate practical applications like scene editing and manipulation, some works extend the framework with semantic masks input for the object-compositional reconstruction rather than the holistic perspective. Though achieving plausible disentanglement, the performance drops significantly when processing the indoor scenes where objects are usually partially observed. We propose RICO to address this by regularizing the unobservable regions for indoor compositional reconstruction. Our key idea is to first regularize the smoothness of the occluded background, which then in turn guides the foreground object reconstruction in unobservable regions based on the object-background relationship. Particularly, we regularize the geometry smoothness of occluded background patches. With the improved background surface, the signed distance function and the reversedly rendered depth of objects can be optimized to bound them within the background range. Extensive experiments show our method outperforms other methods on synthetic and real-world indoor scenes and prove the effectiveness of proposed regularizations. The code is available at https://github.com/kyleleey/RICO.

Physical adversarial attacks against deep neural networks (DNNs) have recently gained increasing attention. The current mainstream physical attacks use printed adversarial patches or camouflage to alter the appearance of the target object. However, these approaches generate conspicuous adversarial patterns that show poor stealthiness. Another physical deployable attack is the optical attack, featuring stealthiness while exhibiting weakly in the daytime with sunlight. In this paper, we propose a novel Reflected Light Attack (RFLA), featuring effective and stealthy in both the digital and physical world, which is implemented by placing the color transparent plastic sheet and a paper cut of a specific shape in front of the mirror to create different colored geometries on the target object. To achieve these goals, we devise a general framework based on the circle to model the reflected light on the target object. Specifically, we optimize a circle (composed of a coordinate and radius) to carry various geometrical shapes determined by the optimized angle. The fill color of the geometry shape and its corresponding transparency are also optimized. We extensively evaluate the effectiveness of RFLA on different datasets and models. Experiment results suggest that the proposed method achieves over 99% success rate on different datasets and models in the digital world. Additionally, we verify the effectiveness of the proposed method in different physical environments by using sunlight or a flashlight.

Thriving underwater applications demand efficient extreme compression technology to realize the transmission of underwater images (UWIs) in very narrow underwater bandwidth. However, existing image compression methods achieve inferior performance on UWIs because they do not consider the characteristics of UWIs: (1) Multifarious underwater styles of color shift and distance-dependent clarity, caused by the unique underwater physical imaging; (2) Massive redundancy between different UWIs, caused by the fact that different UWIs contain several common ocean objects, which have plenty of similarities in structures and semantics. To remove redundancy among UWIs, we first construct an exhaustive underwater multi-scale feature dictionary to provide coarse-to-fine reference features for UWI compression. Subsequently, an extreme UWI compression network with reference to the feature dictionary (RFD-ECNet) is creatively proposed, which utilizes feature match and reference feature variant to significantly remove redundancy among UWIs. To align the multifarious underwater styles and improve the accuracy of feature match, an underwater style normalized block (USNB) is proposed, which utilizes underwater physical priors extracted from the underwater physical imaging model to normalize the underwater styles of dictionary features toward the input. Moreover, a reference feature variant module (RFVM) is designed to adaptively morph the reference features, improving the similarity between the reference and input features. Experimental results on four UWI datasets show that our RFD-ECNet is the first work that achieves a significant BD-rate saving of 31% over the most advanced VVC.

Dynamic imaging involves the recovery of a time-varying 2D or 3D object at each time instant using its undersampled measurements. In particular, in dynamic tomography, only a single projection at a single view angle may be available at a time, making the problem severely ill-posed. In this work, we propose an approach, RED-PSM, which combines for the first time two powerful techniques to address this challenging imaging problem. The first, are partially separable models, which have been used to introduce a low-rank prior for the spatio-temporal object. The second is the recent Regularization by Denoising (RED), which provides a flexible framework to exploit the impressive performance of state-of-the-art image denoising algorithms, for various inverse problems. We propose a partially separable objective with RED and an optimization scheme with variable splitting and ADMM. Our objective is proved to converge to a value corresponding to a stationary point satisfying the first-order optimality conditions. Convergence is accelerated by a particular projection-domain-based initialization. We demonstrate the performance and computational improvements of our proposed RED-PSM with a learned image denoiser by comparing it to a recent deep-prior-based method TD-DIP. Although the emphasis is on dynamic tomography, we also demonstrate the performance advantages of RED-PSM in a dynamic cardiac MRI setting.

Machine learning models are known to be susceptible to adversarial perturbation. One famous attack is the adversarial patch, a particularly crafted sticker that makes the model mispredict the object it is placed on. This attack presents a critical threat to cyber-physical systems that rely on cameras such as autonomous cars. Despite the significance of the problem, conducting research in this setting has been difficult; evaluating attacks and defenses in the real world is exceptionally costly while synthetic data are unrealistic. In this work, we propose the REAP (REalistic Adversarial Patch) benchmark, a digital benchmark that enables the evaluations on real images under real-world conditions. Built on top of the Mapillary Vistas dataset, our benchmark contains over 14,000 traffic signs. Each sign is augmented with geometric and lighting transformations for applying a digitally generated patch realistically onto the sign. Using our benchmark, we perform the first large-scale assessments of adversarial patch attacks under realistic conditions. Our experiments suggest that patch attacks may present a smaller threat than previously believed and that the success rate of an attack on simpler digital simulations is not predictive of its actual effectiveness in practice. Our benchmark is released publicly at https://github.com/wagner-group/reap-benchmark.

In this paper, we introduce a novel approach to novel object captioning which employs relative contrastive learning to learn visual and semantic alignment. Our approach maximizes compatibility between regions and object tags in a contrastive manner. To set up a proper contrastive learning objective, for each image, we augment tags by leveraging the relative nature of positive and negative pairs obtained from foundation models such as CLIP. We then use the rank of each augmented tag in a list as a relative relevance label to contrast each top-ranked tag with a set of lower-ranked tags. This learning objective encourages the top-ranked tags to be more compatible with their image and text context than lower-ranked tags, thus improving the discriminative ability of the learned multi-modality representation. We evaluate our approach on two datasets and show that our proposed RCA-NOC approach outperforms state-of-the-art methods by a large margin, demonstrating its effectiveness in improving vision-language representation for novel object captioning.

We propose RANA, a relightable and articulated neural avatar for the photorealistic synthesis of humans under arbitrary viewpoints, body poses, and lighting. We only require a short video clip of the person to create the avatar and assume no knowledge about the lighting environment. We present a novel framework to model humans while disentangling their geometry, texture, and also lighting environment from monocular RGB videos. To simplify this otherwise ill-posed task we first estimate the coarse geometry and texture of the person via SMPL+D model fitting and then learn an articulated neural representation for photorealistic image generation. RANA first generates the normal and albedo maps of the person in any given target body pose and then uses spherical harmonics lighting to generate the shaded image in the target lighting environment. We also propose to pretrain RANA using synthetic images and demonstrate that it leads to better disentanglement between geometry and texture while also improving robustness to novel body poses. Finally, we also present a new photorealistic synthetic dataset, Relighting Humans, to quantitatively evaluate the performance of the proposed approach.

Dense 3D reconstruction and ego-motion estimation are key challenges in autonomous driving and robotics. Compared to the complex, multi-modal systems deployed today, multi-camera systems provide a simpler, low-cost alternative. However, camera-based 3D reconstruction of complex dynamic scenes has proven extremely difficult, as existing solutions often produce incomplete or incoherent results. We propose R3D3, a multi-camera system for dense 3D reconstruction and ego-motion estimation. Our approach iterates between geometric estimation that exploits spatial-temporal information from multiple cameras, and monocular depth refinement. We integrate multi-camera feature correlation and dense bundle adjustment operators that yield robust geometric depth and pose estimates. To improve reconstruction where geometric depth is unreliable, e.g. for moving objects or low-textured regions, we introduce learnable scene priors via a depth refinement network. We show that this design enables a dense, consistent 3D reconstruction of challenging, dynamic outdoor environments. Consequently, we achieve state-of-the-art dense depth prediction on the DDAD and NuScenes benchmarks.

Predicting the future motion of dynamic agents is of paramount importance to ensuring safety and assessing risks in motion planning for autonomous robots. In this study, we propose a two-stage motion prediction method, called R-Pred, designed to effectively utilize both scene and interaction context using a cascade of the initial trajectory proposal and trajectory refinement networks. The initial trajectory proposal network produces M trajectory proposals corresponding to the M modes of the future trajectory distribution. The trajectory refinement network enhances each of the M proposals using 1) tube-query scene attention (TQSA) and 2) proposal-level interaction attention (PIA) mechanisms. TQSA uses tube-queries to aggregate local scene context features pooled from proximity around trajectory proposals of interest. PIA further enhances the trajectory proposals by modeling inter-agent interactions using a group of trajectory proposals selected by their distances from neighboring agents. Our experiments conducted on Argoverse and nuScenes datasets demonstrate that the proposed refinement network provides significant performance improvements compared to the single-stage baseline and that R-Pred achieves state-of-the-art performance in some categories of the benchmarks.

Category-level 6DoF object pose estimation intends to estimate the rotation, translation, and size of unseen objects. Many previous works use point clouds as a pre-learned shape prior to overcome intra-category variability. The shape prior is deformed to reconstruct instances' point clouds in canonical space and to build dense 3D-3D correspondences between the observed and reconstructed point clouds. However, the pre-learned shape prior is not jointly optimized with estimation networks, and they are trained with a surrogate objective. We propose a novel 6D pose estimation network, named Query6DoF, based on a series of category-specific sparse queries that represent the prior shape. Each query represents a shape component, and these queries are learnable embeddings that can be optimized together with the estimation network according to the point cloud reconstruction loss, the normalized object coordinate loss, and the 6d pose estimation loss. Query6DoF adopts a deformation-and-matching paradigm with attention, where the queries dynamically extract features from regions of interest using the attention mechanism and then directly regress results. Furthermore, Query6DoF reduces computation overhead through the sparseness of the queries and the incorporation of a lightweight global information injection block. With the aforementioned design, Query6DoF achieves state-of-the-art (SOTA) pose estimation performance on the NOCS datasets. The source code and models are available at https://github.com/hustvl/Query6DoF.

3D instance segmentation aims to predict a set of object instances in a scene and represent them as binary foreground masks with corresponding semantic labels. However, object instances are diverse in shape and category,and point clouds are usually sparse, unordered, and irregular, which leads to a query sampling dilemma. Besides,noise background queries interfere with proper scene perception and accurate instance segmentation. To address the above issues, we propose a Query Refinement Transformer termed QueryFormer. The key to our approach is to exploit a query initialization module to optimize the initialization process for the query distribution with a high coverage and low repetition rate. Additionally, we design an affiliated transformer decoder that suppresses the interference of noise background queries and helps the foreground queries focus on instance discriminative parts to predict final segmentation results. Extensive experiments on ScanNetV2 and S3DIS datasets show that our QueryFormer can surpass state-of-the-art 3D instance segmentation methods.

Deepfake has recently raised a plethora of societal concerns over its possible security threats and dissemination of fake information. Much research on deepfake detection has been undertaken. However, detecting low quality as well as simultaneously detecting different qualities of deepfakes still remains a grave challenge. Most SOTA approaches are limited by using a single specific model for detecting certain deepfake video quality type. When constructing multiple models with prior information about video quality, this kind of strategy incurs significant computational cost, as well as model and training data overhead. Further, it cannot be scalable and practical to deploy in real-world settings. In this work, we propose a universal intra-model collaborative learning framework to enable the effective and simultaneous detection of different quality of deepfakes. That is, our approach is the quality-agnostic deepfake detection method, dubbed QAD . In particular, by observing the upper bound of general error expectation, we maximize the dependency between intermediate representations of images from different quality levels via Hilbert-Schmidt Independence Criterion. In addition, an Adversarial Weight Perturbation module is carefully devised to enable the model to be more robust against image corruption while boosting the overall model's performance. Extensive experiments over seven popular deepfake datasets demonstrate the superiority of our QAD model over prior SOTA benchmarks.

Models pre-trained on large datasets such as ImageNet provide the de-facto standard for transfer learning, with both supervised and self-supervised approaches proving effective. However, emerging evidence suggests that any single pre-trained feature will not perform well on diverse downstream tasks. Each pre-training strategy encodes a certain inductive bias, which may suit some downstream tasks but not others. Notably, the augmentations used in both supervised and self-supervised training lead to features with high invariance to spatial and appearance transformations. This renders them sub-optimal for tasks that demand sensitivity to these factors. In this paper we develop a feature that better supports diverse downstream tasks by providing a diverse set of sensitivities and invariances. In particular, we are inspired by Quality-Diversity in evolution, to define a pre-training objective that requires high quality yet diverse features -- where diversity is defined in terms of transformation (in)variances. Our framework plugs in to both supervised and self-supervised pre-training, and produces a small ensemble of features. We further show how downstream tasks can easily and efficiently select their preferred (in)variances. Both empirical and theoretical analysis show the efficacy of our representation and transfer learning approach for diverse downstream tasks.