ICCV 2023 论文列表

In this paper we present an approach for localizing steps of procedural activities in narrated how-to videos. To deal with the scarcity of labeled data at scale, we source the step descriptions from a language knowledge base (wikiHow) containing instructional articles for a large variety of procedural tasks. Without any form of manual supervision, our model learns to temporally ground the steps of procedural articles in how-to videos by matching three modalities: frames, narrations, and step descriptions. Specifically, our method aligns steps to video by fusing information from two distinct pathways: i) direct alignment of step descriptions to frames, ii) indirect alignment obtained by composing steps-to-narrations with narrations-to-video correspondences. Notably, our approach performs global temporal grounding of all steps in an article at once by exploiting order information, and is trained with step pseudo-labels which are iteratively refined and aggressively filtered. In order to validate our model we introduce a new benchmark -- HT-Step -- obtained by manually annotating a 124-hour subset of HowTo100M with steps sourced from wikiHow articles. Experiments on this benchmark as well as zero-shot evaluations on CrossTask demonstrate that our multi-modality alignment yields dramatic gains over several baselines and prior works. Finally, we show that our inner module for matching narration-to-video outperforms by a large margin the state of the art on the HTM-Align narration-video alignment benchmark.

Existing text-to-image generation approaches have set high standards for photorealism and text-image correspondence, largely benefiting from web-scale text-image datasets, which can include up to 5 billion pairs. However, text-to-image generation models trained on domain-specific datasets, such as urban scenes, medical images, and faces, still suffer from low text-image correspondence due to the lack of text-image pairs. Additionally, collecting billions of text-image pairs for a specific domain can be time-consuming and costly. Thus, ensuring high text-image correspondence without relying on web-scale text-image datasets remains a challenging task. In this paper, we present a novel approach for enhancing text-image correspondence by leveraging available semantic layouts. Specifically, we propose a Gaussian-categorical diffusion process that simultaneously generates both images and corresponding layout pairs. Our experiments reveal that we can guide text-to-image generation models to be aware of the semantics of different image regions, by training the model to generate semantic labels for each pixel. We demonstrate that our approach achieves higher text-image correspondence compared to existing text-to-image generation approaches in the Multi-Modal CelebA-HQ and the Cityscapes dataset, where text-image pairs are scarce.

Sparse-view computed tomography (CT)---using a small number of projections for tomographic reconstruction---enables much lower radiation dose to patients and accelerated data acquisition. The reconstructed images, however, suffer from strong artifacts, greatly limiting their diagnostic value. Current trends for sparse-view CT turn to the raw data for better information recovery. The resultant dual-domain methods, nonetheless, suffer from secondary artifacts, especially in ultra-sparse view scenarios, and their generalization to other scanners/protocols is greatly limited. A crucial question arises: have the image post-processing methods reached the limit? Our answer is not yet. In this paper, we stick to image post-processing methods due to great flexibility and propose global representation (GloRe) distillation framework for sparse-view CT, termed GloReDi. First, we propose to learn GloRe with Fourier convolution, so each element in GloRe has an image-wide receptive field. Second, unlike methods that only use the full-view images for supervision, we propose to distill GloRe from intermediate-view reconstructed images that are readily available but not explored in previous literature. The success of GloRe distillation is attributed to two key components: representation directional distillation to align the GloRe directions, and band-pass-specific contrastive distillation to gain clinically important details. Extensive experiments demonstrate the superiority of the proposed GloReDi over the state-of-the-art methods, including dual-domain ones. The source code is available at https://github.com/longzilicart/GloReDi.

Conventional multi-label classification (MLC) methods assume that all samples are fully labeled and identically distributed. Unfortunately, this assumption is unrealistic in large-scale MLC data that has long-tailed (LT) distribution and partial labels (PL). To address the problem, we introduce a novel task, Partial labeling and Long-Tailed Multi-Label Classification (PLT-MLC), to jointly consider the above two imperfect learning environments. Not surprisingly, we find that most LT-MLC and PL-MLC approaches fail to solve the PLT-MLC, resulting in significant performance degradation on the two proposed PLT-MLC benchmarks. Therefore, we propose an end-to-end learning framework: COrrection -> ModificatIon -> balanCe, abbreviated as COMC. Our bootstrapping philosophy is to simultaneously correct the missing labels (Correction) with convinced prediction confidence over a class-aware threshold and to learn from these recall labels during training. We next propose a novel multi-focal modifier loss that simultaneously addresses head-tail imbalance and positive-negative imbalance to adaptively modify the attention to different samples (Modification) under the LT class distribution. We also develop a balanced training strategy by distilling the model's learning effect from head and tail samples, and thus design the balanced classifier (Balance) conditioned on the head and tail learning effect to maintain a stable performance. Our experimental study shows that the proposed method significantly outperforms the general MLC, LT-MLC and ML-MLC methods in terms of effectiveness and robustness on our newly created PLT-MLC datasets.

We are concerned with a challenging scenario in unpaired multiview video learning. In this case, the model aims to learn comprehensive multiview representations while the cross-view semantic information exhibits variations. We propose Semantics-based Unpaired Multiview Learning (SUM-L) to tackle this unpaired multiview learning problem. The key idea is to build cross-view pseudo-pairs and do view-invariant alignment by leveraging the semantic information of videos. To facilitate the data efficiency of multiview learning, we further perform video-text alignment for first-person and third-person videos, to fully leverage the semantic knowledge to improve video representations. Extensive experiments on multiple benchmark datasets verify the effectiveness of our framework. Our method also outperforms multiple existing view-alignment methods, under the more challenging scenario than typical paired or unpaired multimodal or multiview learning. Our code is available at https://github.com/wqtwjt1996/SUM-L.

Semi-Supervised Temporal Action Localization (SS-TAL) aims to improve the generalization ability of action detectors with large-scale unlabeled videos. Albeit the recent advancement, one of the major challenges still remains: noisy pseudo labels hinder efficient learning on abundant unlabeled videos, embodied as location biases and category errors. In this paper, we dive deep into such an important but understudied dilemma. To this end, we propose a unified framework, termed Noisy Pseudo-Label Learning, to handle both location biases and category errors. Specifically, our method is featured with (1) Noisy Label Ranking to rank pseudo labels based on the semantic confidence and boundary reliability, (2) Noisy Label Filtering to address the class-imbalance problem of pseudo labels caused by category errors, (3) Noisy Label Learning to penalize inconsistent boundary predictions to achieve noise-tolerant learning for heavy location biases. As a result, our method could effectively handle the label noise problem and improve the utilization of a large amount of unlabeled videos. Extensive experiments on THUMOS14 and ActivityNet v1.3 demonstrate the effectiveness of our method. The code is available at github.com/kunnxia/NPL.

Pseudo-Labeling (PL) is a critical approach in semi-supervised 3D object detection (SSOD). In PL, delicately selected pseudo-labels, generated by the teacher model, are provided for the student model to supervise the semi-supervised detection framework. However, such a paradigm may introduce misclassified labels or loose localized box predictions, resulting in a sub-optimal solution of detection performance. In this paper, we take PL from a noisy learning perspective: instead of directly applying vanilla pseudo-labels, we design a noise-resistant instance supervision module for better generalization. Specifically, we soften the classification targets by considering both the quality of pseudo labels and the network learning ability, and convert the regression task into a probabilistic modeling problem. Besides, considering that self-supervised learning works in the absence of labels, we incorporate dense pixel-wise feature consistency constraints to eliminate the negative impact of noisy labels. To this end, we propose NoiseDet, a simple yet effective framework for semi-supervised 3D object detection. Extensive experiments on competitive ONCE and Waymo benchmarks demonstrate that our method outperforms current semi-supervised approaches by a large margin. Notably, our NoiseDet achieves state-of-the-art performance under various dataset scales on ONCE dataset. For example, NoiseDet improves its NoiseyStudent baseline from 55.5 mAP to 58.0 mAP, and further reaches 60.2 mAP with enhanced pseudo-label generation. Code will be available at https://github.com/zehuichen123/NoiseDet.

Contrastive learning has become an important tool in learning representations from unlabeled data mainly relying on the idea of minimizing distance between positive data pairs, e.g., views from the same images, and maximizing distance between negative data pairs, e.g., views from different images. This paper proposes a new variation of the contrastive learning objective, Group Ordering Constraints (GroCo), that leverages the idea of sorting the distances of positive and negative pairs and computing the respective loss based on how many positive pairs have a larger distance than the negative pairs, and thus are not ordered correctly. To this end, the GroCo loss is based on differentiable sorting networks, which enable training with sorting supervision by matching a differentiable permutation matrix, which is produced by sorting a given set of scores, to a respective ground truth permutation matrix. Applying this idea to groupwise pre-ordered inputs of multiple positive and negative pairs allows introducing the GroCo loss with implicit emphasis on strong positives and negatives, leading to better optimization of the local neighborhood. We evaluate the proposed formulation on various self-supervised learning benchmarks and show that it not only leads to improved results compared to vanilla contrastive learning but also shows competitive performance to comparable methods in linear probing and outperforms current methods in k-NN performance.

Implicit neural rendering, using signed distance function (SDF) representation with geometric priors like depth or surface normal, has made impressive strides in the surface reconstruction of large-scale scenes. However, applying this method to reconstruct a room-level scene from images may miss structures in low-intensity areas and/or small, thin objects. We have conducted experiments on three datasets to identify limitations of the original color rendering loss and priors-embedded SDF scene representation.Our findings show that the color rendering loss creates an optimization bias against low-intensity areas, resulting in gradient vanishing and leaving these areas unoptimized. To address this issue, we propose a feature-based color rendering loss that utilizes non-zero feature values to bring back optimization signals. Additionally, the SDF representation can be influenced by objects along a ray path, disrupting the monotonic change of SDF values when a single object is present. Accordingly, we explore using the occupancy representation, which encodes each point separately and is unaffected by objects along a querying ray. Our experimental results demonstrate that the joint forces of the feature-based rendering loss and Occ-SDF hybrid representation scheme can provide high-quality reconstruction results, especially in challenging room-level scenarios. The code is available at https://github.com/shawLyu/Occ-SDF_Hybrid.

Latest methods represent shapes with open surfaces using unsigned distance functions (UDFs). They train neural networks to learn UDFs and reconstruct surfaces with the gradients around the zero level set of the UDF. However, the differential networks struggle from learning the zero level set where the UDF is not differentiable, which leads to large errors on unsigned distances and gradients around the zero level set, resulting in highly fragmented and discontinuous surfaces. To resolve this problem, we propose to learn a more continuous zero level set in UDFs with level set projections. Our insight is to guide the learning of zero level set using the rest non-zero level sets via a projection procedure. Our idea is inspired from the observations that the non-zero level sets are much smoother and more continuous than the zero level set. We pull the non-zero level sets onto the zero level set with gradient constraints which align gradients over different level sets and correct unsigned distance errors on the zero level set, leading to a smoother and more continuous unsigned distance field. We conduct comprehensive experiments in surface reconstruction for point clouds, real scans or depth maps, and further explore the performance in unsupervised point cloud upsampling and unsupervised point normal estimation with the learned UDF, which demonstrate our non-trivial improvements over the state-of-the-art methods. Code is available at https://github.com/junshengzhou/LevelSetUDF.

Vision-and-language navigation (VLN) requires an embodied agent to navigate in realistic 3D environments using natural language instructions. Existing VLN methods suffer from training on small-scale environments or unreasonable path-instruction datasets, limiting the generalization to unseen environments. There are massive house tour videos on YouTube, providing abundant real navigation experiences and layout information. However, these videos have not been explored for VLN before. In this paper, we propose to learn an agent from these videos by creating a large-scale dataset which comprises reasonable path-instruction pairs from house tour videos and pre-training the agent on it. To achieve this, we have to tackle the challenges of automatically constructing path-instruction pairs and exploiting real layout knowledge from raw and unlabeled videos. To address these, we first leverage an entropy-based method to construct the nodes of a path trajectory. Then, we propose an action-aware generator for generating instructions from unlabeled trajectories. Last, we devise a trajectory judgment pretext task to encourage the agent to mine the layout knowledge. Experimental results show that our method achieves state-of-the-art performance on two popular benchmarks (R2R and REVERIE). Code is available at https://github.com/JeremyLinky/YouTube-VLN.

Auto-Regressive (AR) models have achieved impressive results in 2D image generation by modeling joint distributions in the grid space. While this approach has been extended to the 3D domain for powerful shape generation, it still has two limitations: expensive computations on volumetric grids and ambiguous auto-regressive order along grid dimensions. To overcome these limitations, we propose the Improved Auto-regressive Model (ImAM) for 3D shape generation, which applies discrete representation learning based on a latent vector instead of volumetric grids. Our approach not only reduces computational costs but also preserves essential geometric details by learning the joint distribution in a more tractable order. Moreover, thanks to the simplicity of our model architecture, we can naturally extend it from unconditional to conditional generation by concatenating various conditioning inputs, such as point clouds, categories, images, and texts. Extensive experiments demonstrate that ImAM can synthesize diverse and faithful shapes of multiple categories, achieving state-of-the-art performance.

Implicit neural representations have shown powerful capacity in modeling real-world 3D scenes, offering superior performance in novel view synthesis. In this paper, we target a more challenging scenario, i.e., joint scene novel view synthesis and editing based on implicit neural scene representations. State-of-the-art methods in this direction typically consider building separate networks for these two tasks (i.e., view synthesis and editing). Thus, the modeling of interactions and correlations between these two tasks is very limited, which, however, is critical for learning high-quality scene representations.To tackle this problem, in this paper, we propose a unified Neural Radiance Field (NeRF) framework to effectively perform joint scene decomposition and composition for modeling real-world scenes. The decomposition aims at learning disentangled 3D representations of different objects and the background, allowing for scene editing, while scene composition models an entire scene representation for novel view synthesis. Specifically, with a two-stage NeRF framework, we learn a coarse stage for predicting a global radiance field as guidance for point sampling, and in the second fine-grained stage, we perform scene decomposition by a novel one-hot object radiance field regularization module and a pseudo supervision via inpainting to handle ambiguous background regions occluded by objects. The decomposed object-level radiance fields are further composed by using activations from the decomposition module. Extensive quantitative and qualitative results show the effectiveness of our method for scene decomposition and composition, outperforming state-of-the-art methods for both novel-view synthesis and editing tasks.

In a video, an object usually appears as the trajectory, i.e., it spans over a few spatial but longer temporal patches, that contains abundant spatiotemporal contexts. However, modern Video-Language BERTs (VDL-BERTs) neglect this trajectory characteristic that they usually follow image-language BERTs (IL-BERTs) to deploy the patch-to-word (P2W) attention that may over-exploit trivial spatial contexts and neglect significant temporal contexts. To amend this, we propose a novel TW-BERT to learn Trajectory-Word alignment by a newly designed trajectory-to-word (T2W) attention for solving video-language tasks. Moreover, previous VDL-BERTs usually uniformly sample a few frames into the model while different trajectories have diverse graininess, i.e., some trajectories span longer frames and some span shorter, and using a few frames will lose certain useful temporal contexts. However, simply sampling more frames will also make pre-training infeasible due to the largely increased training burdens. To alleviate the problem, during the fine-tuning stage, we insert a novel Hierarchical Frame-Selector (HFS) module into the video encoder. HFS gradually selects the suitable frames conditioned on the text context for the later cross-modal encoder to learn better trajectory-word alignments. By the proposed T2W attention and HFS, our TW-BERT achieves SOTA performances on text-to-video retrieval tasks, and comparable performances on video question-answering tasks with some VDL-BERTs trained on much more data. The code will be available in the supplementary material.

Current 6D pose estimation methods focus on handling objects that are previously trained, which limits their applications in real dynamic world. To this end, we propose a geometry correspondence-based framework, termed GCPose, to estimate 6D pose of arbitrary unseen objects without any re-training. Specifically, the proposed method draws the idea from point cloud registration and resorts to object-agnostic geometry features to establish the 3D-3D correspondences between the object-scene point cloud and object-model point cloud. Then the 6D pose parameters are solved by a least-squares fitting algorithm. Taking the symmetry properties of objects into consideration, we design a symmetry-aware matching loss to facilitate the learning of dense point-wise geometry features and improve the performance considerably. Moreover, we introduce an online training data generation with special data augmentation and normalization to empower the network to learn diverse geometry prior. With training on synthetic objects from ShapeNet, our method outperforms previous approaches for unseen object pose estimation by a large margin on T-LESS, LINEMOD, Occluded-LINEMOD, and TUD-L datasets. Code is available at https://github.com/hikvision-research/GCPose.

Prototypical part network (ProtoPNet) methods have been designed to achieve interpretable classification by associating predictions with a set of training prototypes, which we refer to as trivial prototypes because they are trained to lie far from the classification boundary in the feature space. Note that it is possible to make an analogy between ProtoPNet and support vector machine (SVM) given that the classification from both methods relies on computing similarity with a set of training points (i.e., trivial prototypes in ProtoPNet, and support vectors in SVM). However, while trivial prototypes are located far from the classification boundary, support vectors are located close to this boundary, and we argue that this discrepancy with the well-established SVM theory can result in ProtoPNet models with inferior classification accuracy. In this paper, we aim to improve the classification of ProtoPNet with a new method to learn support prototypes that lie near the classification boundary in the feature space, as suggested by the SVM theory. In addition, we target the improvement of classification results with a new model, named ST-ProtoPNet, which exploits our support prototypes and the trivial prototypes to provide more effective classification. Experimental results on CUB-200-2011, Stanford Cars, and Stanford Dogs datasets demonstrate that ST-ProtoPNet achieves state-of-the-art classification accuracy and interpretability results. We also show that the proposed support prototypes tend to be better localised in the object of interest rather than in the background region. Code is available at https://github.com/cwangrun/ST-ProtoPNet.

In instance image retrieval, considering local spatial information within an image has proven effective to boost retrieval performance, as demonstrated by local visual descriptor based geometric verification. Nevertheless, it will be highly valuable to make ordinary global image representations spatial-context-aware because global representation based image retrieval is appealing thanks to its algorithmic simplicity, low memory cost, and being friendly to sophisticated data structures. To this end, we propose a novel feature learning framework for instance image retrieval, which embeds local spatial context information into the learned global feature representations. Specifically, in parallel to the visual feature branch in a CNN backbone, we design a spatial context branch that consists of two modules called online token learning and distance encoding. For each local descriptor learned in CNN, the former module is used to indicate the types of its surrounding descriptors, while their spatial distribution information is captured by the latter module. After that, the visual feature branch and the spatial context branch are fused to produce a single global feature representation per image. As experimentally demonstrated, with the spatial-context-aware characteristic, we can well improve the performance of global representation based image retrieval while maintaining all of its appealing properties. Our code is available at https://github.com/Zy-Zhang/SpCa

Shape primitives decomposition has been an important and long-standing task in 3D shape analysis. Prior arts heavily rely on 3D point clouds or voxel data for shape primitives extraction, which are less practical in real-world scenarios. This paper proposes to learn shape primitives from multi-view images by introducing implicit surface rendering. It is challenging since implicit shapes have a high degree of freedom, which violates the simplicity property of shape primitives. In this work, a novel regularization term named Implicit Convexity Regularization (ICR) imposed on implicit primitive learning is proposed to tackle this problem. We start with the convexity definition of general 3D shapes, and then derive the equivalent expression for implicit shapes represented by signed distance functions (SDFs). Further, instead of directly constraining the output SDF values which cause unstable optimization, we alternatively impose constraint on second order directional derivatives on line segments inside the shapes, which proves to be a tighter condition for 3D convexity. Implicit primitives constrained by the proposed ICR are combined into a whole object via softmax-weighted-sum operation over all primitive SDFs. Experiments on synthetic and real-world datasets show that our method is able to decompose objects into simple and reasonable shape primitives without the need of segmentation labels or 3D data. Code and data is publicly available in https://github.com/seanywang0408/ICR.

In this paper, we address the problem of generalized category discovery (GCD), i.e., given a set of images where part of them are labelled and the rest are not, the task is to automatically cluster the images in the unlabelled data, leveraging the information from the labelled data, while the unlabelled data contain images from the labelled classes and also new ones. GCD is similar to semi-supervised learning (SSL) but is more realistic and challenging, as SSL assumes all the unlabelled images are from the same classes as the labelled ones. We also do not assume the class number in the unlabelled data is known a-priori, making the GCD problem even harder. To tackle the problem of GCD without knowing the class number, we propose an EM-like framework that alternates between representation learning and class number estimation. We propose a semi-supervised variant of the Gaussian Mixture Model (GMM) with a stochastic splitting and merging mechanism to dynamically determine the prototypes by examining the cluster compactness and separability. With these prototypes, we leverage prototypical contrastive learning for representation learning on the partially labelled data subject to the constraints imposed by the labelled data. Our framework alternates between these two steps until convergence. The cluster assignment for an unlabelled instance can then be retrieved by identifying its nearest prototype. We comprehensively evaluate our framework on both generic image classification datasets and challenging fine-grained object recognition datasets, achieving state-of-the-art performance.