ICCV 2023 论文列表

Multi-turn textual feedback-based fashion image retrieval focuses on a real-world setting, where users can iteratively provide information to refine retrieval results until they find an item that fits all their requirements. In this work, we present a novel memory-based method, called FashionNTM, for such a multi-turn system. Our framework incorporates a new Cascaded Memory Neural Turing Machine (CM-NTM) approach for implicit state management, thereby learning to integrate information across all past turns to retrieve new images, for a given turn. Unlike vanilla Neural Turing Machine (NTM), our CM-NTM operates on multiple inputs, which interact with their respective memories via individual read and write heads, to learn complex relationships. Extensive evaluation results show that our proposed method outperforms the previous state-of-the-art algorithm by 50.5%, on Multi-turn FashionIQ -- the only existing multi-turn fashion dataset currently, in addition to having a relative improvement of 12.6% on Multi-turn Shoes -- an extension of the single-turn Shoes dataset that we created in this work. Further analysis of the model in a real-world interactive setting demonstrates two important capabilities of our model -- memory retention across turns, and agnosticity to turn order for non-contradictory feedback. Finally, user study results show that images retrieved by FashionNTM were favored by 83.1% over other multi-turn models.

Automatic 3D content creation has achieved rapid progress recently due to the availability of pre-trained, large language models and image diffusion models, forming the emerging topic of text-to-3D content creation. Existing text-to-3D methods commonly use implicit scene representations, which couple the geometry and appearance via volume rendering and are suboptimal in terms of recovering finer geometries and achieving photorealistic rendering; consequently, they are less effective for generating high-quality 3D assets. In this work, we propose a new method of Fantasia3D for high-quality text-to-3D content creation. Key to Fantasia3D is the disentangled modeling and learning of geometry and appearance. For geometry learning, we rely on a hybrid scene representation, and propose to encode surface normal extracted from the representation as the input of the image diffusion model. For appearance modeling, we introduce the spatially varying bidirectional reflectance distribution function (BRDF) into the text-to-3D task, and learn the surface material for photorealistic rendering of the generated surface. Our disentangled framework is more compatible with popular graphics engines, supporting relighting, editing, and physical simulation of the generated 3D assets. We conduct thorough experiments that show the advantages of our method over existing ones under different text-to-3D task settings. Project page and source codes: https://fantasia3d.github.io/.

Fine-grained leaf image retrieval (FGLIR) aims to search similar leaf images in subspecies level which involves very high interclass visual similarity and accordingly poses great challenges to leaf image description. In this study, we introduce a new concept, named fan-beam binarization difference projection (FB-BDP) to address this challenging issue. It is designed based on the theory of fan-beam projection (FBP) which is a mathematical tool originally used for computed tomographic reconstruction of objects and has the merits of capturing the inner structure information of objects in multiple directions and excellent ability to suppress image noise. However, few studies have been made to apply FBP to the description of texture patterns. Rather than calculating ray integrals over the whole object area, FB-BDP restricts its ray integrals calculated over local patches to guarantee the locality of the extracted features. By binarizing the intensity-differences between the off-center and center rays, FB-BDP enable its ray integrals insensitive to illumination change and more discriminative in the characterization of texture patterns. In additional, due to inheriting the merits of FBP, the proposed FB-BDP is superior over the existing local image descriptors by its invariance to scaling transformation, robustness to noise, and strong ability to capture direction and structure texture patterns. The results of extensive experiments on FGLIR show its higher retrieval accuracy over the benchmark methods, promising generalization power and strong complementarity to deep features.

As recent advances in Neural Radiance Fields (NeRF) have enabled high-fidelity 3D face reconstruction and novel view synthesis, its manipulation also became an essential task in 3D vision. However, existing manipulation methods require extensive human labor, such as a user-provided semantic mask and manual attribute search unsuitable for non-expert users. Instead, our approach is designed to require a single text to manipulate a face reconstructed with NeRF. To do so, we first train a scene manipulator, a latent code-conditional deformable NeRF, over a dynamic scene to control a face deformation using the latent code. However, representing a scene deformation with a single latent code is unfavorable for compositing local deformations observed in different instances. As so, our proposed Position-conditional Anchor Compositor (PAC) learns to represent a manipulated scene with spatially varying latent codes. Their renderings with the scene manipulator are then optimized to yield high cosine similarity to a target text in CLIP embedding space for text-driven manipulation. To the best of our knowledge, our approach is the first to address the text-driven manipulation of a face reconstructed with NeRF. Extensive results, comparisons, and ablation studies demonstrate the effectiveness of our approach.

Face clustering is a method for unlabeled image annotation and has attracted increasing attention. Existing methods have made significant breakthroughs by introducing Graph Convolutional Networks (GCNs) on the affinity graph. However, such graphs will contain many vertex pairs with inconsistent similarities and labels, thus degrading the model's performance. There are already relevant efforts for this problem, but the information about features needs to be mined further. In this paper, we define a new concept called confidence edge and guide the construction of graphs. Furthermore, a novel confidence-GCN is proposed to cluster face images by deriving more confidence edges. Firstly, Local Information Fusion is advanced to obtain a more accurate similarity metric by considering the neighbors of vertices. Then Unsupervised Neighbor Determination is used to discard low-quality edges based on similarity differences. Moreover, we elaborate that the remaining edges retain the most beneficial information to demonstrate the validity. At last, the confidence-GCN takes the graph as the input and fully uses the confidence edges to complete the clustering. Experiments show that our method outperforms existing methods on the face and person datasets to achieve state-of-the-art. At the same time, comparable results are obtained on the fashion dataset.

Multi-modality fusion and multi-task learning are becoming trendy in 3D autonomous driving scenario, considering robust prediction and computation budget. However, naively extending the existing framework to the domain of multi-modality multi-task learning remains ineffective and even poisonous due to the notorious modality bias and task conflict. Previous works manually coordinate the learning framework with empirical knowledge, which may lead to sub-optima. To mitigate the issue, we propose a novel yet simple multi-level gradient calibration learning framework across tasks and modalities during optimization. Specifically, the gradients, produced by the task heads and used to update the shared backbone, will be calibrated at the backbone's last layer to alleviate the task conflict. Before the calibrated gradients are further propagated to the modality branches of the backbone, their magnitudes will be calibrated again to the same level, ensuring the downstream tasks pay balanced attention to different modalities. Experiments on large-scale benchmark nuScenes demonstrate the effectiveness of the proposed method, eg, an absolute 14.4% mIoU improvement on map segmentation and 1.4% mAP improvement on 3D detection, advancing the application of 3D autonomous driving in the domain of multi-modality fusion and multi-task learning. We also discuss the links between modalities and tasks.

Under-display camera (UDC) systems remove the screen notch for bezel-free displays and provide a better interactive experience. The main challenge is that the pixel array of light-emitting diodes used for display diffracts and attenuates the incident light, leading to complex degradation. Existing models eliminate spatial diffraction by maximizing model capacity through complex design and ignore the periodic distribution of diffraction in the frequency domain, which prevents these approaches from satisfactory results. In this paper, we introduce a new perspective to handle various diffraction in UDC images by jointly exploring the feature restoration in the frequency and spatial domains, and present a Frequency and Spatial Interactive Learning Network (FSI). It consists of a series of well-designed Frequency-Spatial Joint (FSJ) modules for feature learning and a color transform module for color enhancement. In particular, in the FSJ module, a frequency learning block uses the Fourier transform to eliminate spectral bias, a spatial learning block uses a multi-distillation structure to supplement the absence of local details, and a dual transfer unit to facilitate the interactive learning between features of different domains. Experimental results demonstrate the superiority of the proposed FSI over state-of-the-art models, through extensive quantitative and qualitative evaluations in three widely-used UDC benchmarks.

Existing skeleton-based action recognition methods typically follow a centralized learning paradigm, which can pose privacy concerns when exposing human-related videos. Federated Learning (FL) has attracted much attention due to its outstanding advantages in privacy-preserving. However, directly applying FL approaches to skeleton videos suffers from unstable training. In this paper, we investigate and discover that the heterogeneous human topology graph structure is the crucial factor hindering training stability. To address this issue, we pioneer a novel Federated Skeleton-based Action Recognition (FSAR) paradigm, which enables the construction of a globally generalized model without accessing local sensitive data. Specifically, we introduce an Adaptive Topology Structure (ATS), separating generalization and personalization by learning a domain-invariant topology shared across clients and a domain-specific topology decoupled from global model aggregation. Furthermore, we explore Multi-grain Knowledge Distillation (MKD) to mitigate the discrepancy between clients and the server caused by distinct updating patterns through aligning shallow block-wise motion features. Extensive experiments on multiple datasets demonstrate that FSAR outperforms state-of-the-art FL-based methods while inherently protecting privacy for skeleton-based action recognition.

This paper is on Few-Shot Object Detection (FSOD), where given a few templates (examples) depicting a novel class (not seen during training), the goal is to detect all of its occurrences within a set of images. From a practical perspective, an FSOD system must fulfil the following desiderata: (a) it must be used as is, without requiring any fine-tuning at test time, (b) it must be able to process an arbitrary number of novel objects concurrently while supporting an arbitrary number of examples from each class and (c) it must achieve accuracy comparable to a closed system. Towards satisfying (a)-(c), in this work, we make the following contributions: We introduce, for the first time, a simple, yet powerful, few-shot detection transformer (FS-DETR) based on visual prompting that can address both desiderata (a) and (b). Our system builds upon the DETR framework, extending it based on two key ideas: (1) feed the provided visual templates of the novel classes as visual prompts during test time, and (2) "stamp" these prompts with pseudo-class embeddings (akin to soft prompting), which are then predicted at the output of the decoder. Importantly, we show that our system is not only more flexible than existing methods, but also, it makes a step towards satisfying desideratum (c). Specifically, it is significantly more accurate than all methods that do not require fine-tuning and even matches and outperforms the current state-of-the-art fine-tuning based methods on the most well-established benchmarks (PASCAL VOC & MSCOCO).

Federated Learning (FL) is a decentralized machine learning paradigm, in which multiple clients collaboratively train neural networks without centralizing their local data, and hence preserve data privacy. However, real-world FL applications usually encounter challenges arising from distribution shifts across the local datasets of individual clients. These shifts may drift the global model aggregation or result in convergence to deflected local optimum. While existing efforts have addressed distribution shifts in the label space, an equally important challenge remains relatively unexplored. This challenge involves situations where the local data of different clients indicate identical label distributions but exhibit divergent feature distributions. This issue can significantly impact the global model performance in the FL framework. In this work, we propose Federated Representation Augmentation (FRAug) to resolve this practical and challenging problem. FRAug optimizes a shared embedding generator to capture client consensus. Its output synthetic embeddings are transformed into client-specific by a locally optimized RTNet to augment the training space of each client. Our empirical evaluation on three public benchmarks and a real-world medical dataset demonstrates the effectiveness of the proposed method, which substantially outperforms the current state-of-the-art FL methods for feature distribution shifts, including PartialFed and FedBN.

Many weakly supervised semantic segmentation (WSSS) methods employ the class activation map (CAM) to generate the initial segmentation results. However, CAM often fails to distinguish the foreground from its co-occurred background (e.g., train and railroad), resulting in inaccurate activation from the background. Previous endeavors address this co-occurrence issue by introducing external supervision and human priors. In this paper, we present a False Positive Rectification (FPR) approach to tackle the co-occurrence problem by leveraging the false positives of CAM. Based on the observation that the CAM-activated regions of absent classes contain class-specific co-occurred background cues, we collect these false positives and utilize them to guide the training of CAM network by proposing a region-level contrast loss and a pixel-level rectification loss. Without introducing any external supervision and human priors, the proposed FPR effectively suppresses wrong activations from the background objects. Extensive experiments on the PASCAL VOC 2012 and MS COCO 2014 demonstrate that FPR brings significant improvements for off-the-shelf methods and achieves state-of-the-art performance. Code is available at https://github.com/mt-cly/FPR.

The quadratic computation complexity of self-attention has been a persistent challenge when applying Transformer models to vision tasks. Linear attention, on the other hand, offers a much more efficient alternative with its linear complexity by approximating the Softmax operation through carefully designed mapping functions. However, current linear attention approaches either suffer from significant performance degradation or introduce additional computation overhead from the mapping functions. In this paper, we propose a novel Focused Linear Attention module to achieve both high efficiency and expressiveness. Specifically, we first analyze the factors contributing to the performance degradation of linear attention from two perspectives: the focus ability and feature diversity. To overcome these limitations, we introduce a simple yet effective mapping function and an efficient rank restoration module to enhance the expressiveness of self-attention while maintaining low computation complexity. Extensive experiments show that our linear attention module is applicable to a variety of advanced vision Transformers, and achieves consistently improved performances on multiple benchmarks. Code is available at https://github.com/LeapLabTHU/FLatten-Transformer.

Face anti-spoofing (FAS) or presentation attack detection is an essential component of face recognition systems deployed in security-critical applications. Existing FAS methods have poor generalizability to unseen spoof types, camera sensors, and environmental conditions. Recently, vision transformer (ViT) models have been shown to be effective for the FAS task due to their ability to capture long-range dependencies among image patches. However, adaptive modules or auxiliary loss functions are often required to adapt pre-trained ViT weights learned on large-scale datasets such as ImageNet. In this work, we first show that initializing ViTs with multimodal (e.g., CLIP) pre-trained weights improves generalizability for the FAS task, which is in line with the zero-shot transfer capabilities of vision-language pre-trained (VLP) models. We then propose a novel approach for robust cross-domain FAS by grounding visual representations with the help of natural language. Specifically, we show that aligning the image representation with an ensemble of class descriptions (based on natural language semantics) improves FAS generalizability in low-data regimes. Finally, we propose a multimodal contrastive learning strategy to boost feature generalization further and bridge the gap between source and target domains. Extensive experiments on three standard protocols demonstrate that our method significantly outperforms the state-of-the-art methods, achieving better zero-shot transfer performance than five-shot transfer of "adaptive ViTs". Code: https://github.com/koushiksrivats/FLIP

Despite the fact that transformer-based models have yielded great success in computer vision tasks, they suffer from the challenge of high computational costs that limits their use on resource-constrained devices. One major reason is that vision transformers have redundant calculations since the self-attention operation generates patches with high similarity at a later stage in the network. Hierarchical architectures have been proposed for vision transformers to alleviate this challenge. However, by shrinking the spatial dimensions to half of the originals with downsampling layers, the challenge is actually overcompensated, as too much information is lost. In this paper, we propose FDViT to improve the hierarchical architecture of the vision transformer by using a flexible downsampling layer that is not limited to integer stride to smoothly reduce the sizes of the middle feature maps. Furthermore, a masked auto-encoder architecture is used to facilitate the training of the proposed flexible downsampling layer and produces informative outputs. Experimental results on benchmark datasets demonstrate that the proposed method can reduce computational costs while increasing classification performance and achieving state-of-the-art results. For example, the proposed FDViT-S model achieves a top-1 accuracy of 81.5%, which is 1.7 percent points higher than the ViT-S model and reduces 39% FLOPs.

Although complex-valued convolutional neural networks (iCNNs) have existed for a while, they lack proper complex-valued image inputs and loss functions. In addition, all their operations are not complex-valued as they have both complex-valued convolutional layers and real-valued fully-connected layers. As a result, they lack an end-to-end flow of complex-valued information, making them inconsistent w.r.t. the claimed operating domain, i.e., complex numbers. Considering these inconsistencies, we propose a complex-valued color model and loss function and turn fully-connected layers into convolutional layers. All these contributions culminate in what we call FCCNs (Fully Complex-valued Convolutional Networks), which take complex-valued images as inputs, perform only complex-valued operations, and have a complex-valued loss function. Thus, our proposed FCCNs have an end-to-end flow of complex-valued information, which lacks in existing iCNNs. Our extensive experiments on five image classification benchmark datasets show that FCCNs consistently perform better than existing iCNNs. Code is available at https://github.com/saurabhya/FCCNs .

The problem of predicting driver attention from the driving perspective is gaining increasing research focus due to its remarkable significance for autonomous driving and assisted driving systems. The driving experience is extremely important for safe driving, a skilled driver is able to effortlessly predict oncoming danger (before it becomes salient) based on the driving experience and quickly pay attention to the corresponding zones. However, the nonobjective driving experience is difficult to model, so a mechanism simulating the driver experience accumulation procedure is absent in existing methods, and the current methods usually follow the technique line of saliency prediction methods to predict driver attention. In this paper, we propose a FeedBack Loop Network (FBLNet), which attempts to model the driving experience accumulation procedure. By over-and-over iterations, FBLNet generates the incremental knowledge that carries rich historically-accumulative and long-term temporal information. The incremental knowledge in our model is like the driving experience of humans. Under the guidance of the incremental knowledge, our model fuses the CNN feature and Transformer feature that are extracted from the input image to predict driver attention. Our model exhibits a solid advantage over existing methods, achieving an outstanding performance improvement on two driver attention benchmark datasets.

View Transformation Module (VTM), where transformations happen between multi-view image features and Bird-Eye-View (BEV) representation, is a crucial step in camera-based BEV perception systems. Currently, the two most prominent VTM paradigms are forward projection and backward projection. Forward projection, represented by Lift-Splat-Shoot, leads to sparsely projected BEV features without post-processing. Backward projection, with BEVFormer being an example, tends to generate false-positive BEV features from incorrect projections due to the lack of utilization on depth. To address the above limitations, we propose a novel forward-backward view transformation module. Our approach compensates for the deficiencies in both existing methods, allowing them to enhance each other to obtain higher quality BEV representations mutually. We instantiate the proposed module with FB-BEV, which achieves a new state-of-the-art result of 62.4% NDS on the nuScenes test set. Code and models are available at https://github.com/NVlabs/FB-BEV

Computer vision datasets frequently contain spurious correlations between task-relevant labels and (easy to learn) latent task-irrelevant attributes (e.g. context). Models trained on such datasets learn "shortcuts" and underperform on bias-conflicting slices of data where the correlation does not hold. In this work, we study the problem of identifying such slices to inform downstream bias mitigation strategies. We propose First Amplify Correlations and Then Slice (FACTS), wherein we first amplify correlations to fit a simple bias-aligned hypothesis via strongly regularized empirical risk minimization. Next, we perform correlation-aware slicing via mixture modeling in bias-aligned feature space to discover underperforming data slices that capture distinct correlations. Despite its simplicity, our method considerably improves over prior work (by as much as 35% precision@10) in correlation bias identification across a range of diverse evaluation settings. Our code is available at https://github.com/yvsriram/FACTS.

Computer vision models have known performance disparities across attributes such as gender and skin tone. This means during tasks such as classification and detection, model performance differs for certain classes based on the demographics of the people in the image. These disparities have been shown to exist, but until now there has not been a unified approach to measure these differences for common use-cases of computer vision models. We present a new benchmark named FACET (FAirness in Computer Vision EvaluaTion), a large, publicly available evaluation set of 32k images for some of the most common vision tasks - image classification, object detection and segmentation. For every image in FACET, we hired expert reviewers to manually annotate person-related attributes such as perceived skin tone and hair type, manually draw bounding boxes and label fine-grained person-related classes such as disk jockey or guitarist. In addition, we use FACET to benchmark state-of-the-art vision models and present a deeper understanding of potential performance disparities and challenges across sensitive demographic attributes. With the exhaustive annotations collected, we probe models using single demographics attributes as well as multiple attributes using an intersectional approach (e.g. hair color and perceived skin tone). Our results show that classification, detection, segmentation, and visual grounding models exhibit performance disparities across demographic attributes and intersections of attributes. These harms suggest that not all people represented in datasets receive fair and equitable treatment in these vision tasks. We hope current and future results using our benchmark will contribute to fairer, more robust vision models. FACET is available publicly at https://facet.metademolab.com