International Journal of Computer Vision最新文献

End-to-end (E2E) training has become the de-facto standard for training modern deep networks, e.g., ConvNets and vision Transformers (ViTs). Typically, a global error signal is generated at the end of a model and back-propagated layer-by-layer to update the parameters. This paper shows that the reliance on back-propagating global errors may not be necessary for deep learning. More precisely, deep networks with a competitive or even better performance can be obtained by purely leveraging locally supervised learning, i.e., splitting a network into gradient-isolated modules and training them with local supervision signals. However, such an extension is non-trivial. Our experimental and theoretical analysis demonstrates that simply training local modules with an E2E objective tends to be short-sighted, collapsing task-relevant information at early layers, and hurting the performance of the full model. To avoid this issue, we propose an information propagation (InfoPro) loss, which encourages local modules to preserve as much useful information as possible, while progressively discarding task-irrelevant information. As InfoPro loss is difficult to compute in its original form, we derive a feasible upper bound as a surrogate optimization objective, yielding a simple but effective algorithm. We evaluate InfoPro extensively with ConvNets and ViTs, based on twelve computer vision benchmarks organized into five tasks (i.e., image/video recognition, semantic/instance segmentation, and object detection). InfoPro exhibits superior efficiency over E2E training in terms of GPU memory footprints, convergence speed, and training data scale. Moreover, InfoPro enables the effective training of more parameter- and computation-efficient models (e.g., much deeper networks), which suffer from inferior performance when trained in E2E. Code: https://github.com/blackfeather-wang/InfoPro-Pytorch.

LiDAR-based 3D object detection is a crucial task for autonomous driving, owing to its accurate object recognition and localization capabilities in the 3D real-world space. However, existing methods heavily rely on time-consuming and laborious large-scale labeled LiDAR data, posing a bottleneck for both performance improvement and practical applications. In this paper, we propose Contrastive Masked AutoEncoders for self-supervised 3D object detection, dubbed as CMAE-3D, which is a promising solution to effectively alleviate label dependency in 3D perception. Specifically, we integrate Contrastive Learning (CL) and Masked AutoEncoders (MAE) into one unified framework to fully utilize the complementary characteristics of global semantic representation and local spatial perception. Furthermore, from the perspective of MAE, we develop the Geometric-Semantic Hybrid Masking (GSHM) to selectively mask representative regions in point clouds with imbalanced foreground-background and uneven density distribution, and design the Multi-scale Latent Feature Reconstruction (MLFR) to capture high-level semantic features while mitigating the redundant reconstruction of low-level details. From the perspective of CL, we present Hierarchical Relational Contrastive Learning (HRCL) to mine rich semantic similarity information while alleviating the issue of negative sample mismatch from both the voxel-level and frame-level. Extensive experiments demonstrate the effectiveness of our pre-training method when applied to multiple mainstream 3D object detectors (SECOND, CenterPoint and PV-RCNN) on three popular datasets (KITTI, Waymo and nuScenes).

Differences in forgery attributes of images generated in CNN-synthesized and image-editing domains are large, and such differences make a unified image forgery detection and localization (IFDL) challenging. To this end, we present a hierarchical fine-grained formulation for IFDL representation learning. Specifically, we first represent forgery attributes of a manipulated image with multiple labels at different levels. Then, we perform fine-grained classification at these levels using the hierarchical dependency between them. As a result, the algorithm is encouraged to learn both comprehensive features and the inherent hierarchical nature of different forgery attributes, thereby improving the IFDL representation. In this work, we propose a Language-guided Hierarchical Fine-grained IFDL, denoted as HiFi-Net++. Specifically, HiFi-Net++ contains four components: multi-branch feature extractor, language-guided forgery localization enhancer, as well as classification and localization modules. Each branch of the multi-branch feature extractor learns to classify forgery attributes at one level, while localization and classification modules segment the pixel-level forgery region and detect image-level forgery, respectively. In addition, the language-guided forgery localization enhancer (LFLE), containing image and text encoders learned by contrastive language-image pre-training (CLIP), is used to further enrich the IFDL representation. LFLE takes specifically designed texts and the given image as multi-modal inputs and then generates the visual embedding and manipulation score maps, which are used to further improve HiFi-Net++ manipulation localization performance. Lastly, we construct a hierarchical fine-grained dataset to facilitate our study. We demonstrate the effectiveness of our method on 8 different benchmarks for both tasks of IFDL and forgery attribute classification. Our source code and dataset can be found: github.com/CHELSEA234/HiFi-IFDL.

Large-scale language-vision pre-training models, such as CLIP, have achieved remarkable results in text-guided image morphing by leveraging several unconditional generative models. However, existing CLIP-guided methods face challenges in achieving photorealistic morphing when adapting the generator from the source to the target domain. Specifically, current guidance methods fail to provide detailed explanations of the morphing regions within the image, leading to misguidance and catastrophic forgetting of the original image’s fidelity. In this paper, we propose a novel approach considering proper regularization losses to overcome these difficulties by addressing the SP dilemma in CLIP guidance. Our approach consists of two key components: (1) a geodesic cosine similarity loss that minimizes inter-modality features (i.e., image and text) in a projected subspace of CLIP space, and (2) a latent regularization loss that minimizes intra-modality features (i.e., image and image) on the image manifold. By replacing the naive directional CLIP loss in a drop-in replacement manner, our method achieves superior morphing results for both images and videos across various benchmarks, including CLIP-inversion.

Image-based virtual try-on aims to synthesize a naturally dressed person image with a clothing image, which revolutionizes online shopping and inspires related topics within image generation, showing both research significance and commercial potential. However, there is a gap between current research progress and commercial applications and an absence of comprehensive overview of this field to accelerate the development. In this survey, we provide a comprehensive analysis of the state-of-the-art techniques and methodologies in aspects of pipeline architecture, person representation and key modules such as try-on indication, clothing warping and try-on stage. We additionally apply CLIP to assess the semantic alignment of try-on results, and evaluate representative methods with uniformly implemented evaluation metrics on the same dataset. In addition to quantitative and qualitative evaluation of current open-source methods, unresolved issues are highlighted and future research directions are prospected to identify key trends and inspire further exploration. The uniformly implemented evaluation metrics, dataset and collected methods will be made public available at https://github.com/little-misfit/Survey-Of-Virtual-Try-On.

Zero-cost proxies are nowadays frequently studied and used to search for neural architectures. They show an impressive ability to predict the performance of architectures by making use of their untrained weights. These techniques allow for immense search speed-ups. So far the joint search for well performing and robust architectures has received much less attention in the field of NAS. Therefore, the main focus of zero-cost proxies is the clean accuracy of architectures, whereas the model robustness should play an evenly important part. In this paper, we analyze the ability of common zero-cost proxies to serve as performance predictors for robustness in the popular NAS-Bench-201 search space. We are interested in the single prediction task for robustness and the joint multi-objective of clean and robust accuracy. We further analyze the feature importance of the proxies and show that predicting the robustness makes the prediction task from existing zero-cost proxies more challenging. As a result, the joint consideration of several proxies becomes necessary to predict a model’s robustness while the clean accuracy can be regressed from a single such feature. Our code is available at https://github.com/jovitalukasik/zcp_eval.

Video synopsis is a technique that condenses a long surveillance video to a short summary. It faces challenges to process objects originally occluding each other in the source video. Previous approaches either treat occlusion objects as a single object, which however reduce compression ratio; or have to separate occlusion objects individually, but destroy interactions between them and yield visual artifacts. This paper presents a novel data structure called Flexible Object Graph (FOG) to handle original occlusions. Our FOG-based video synopsis approach can manipulate each object flexibly while preserving the original occlusions between them, achieving high synopsis ratio while maintaining interactions of objects. A challenging issue that comes with the introduction of FOG is that FOG may contain circulations that yield conflicts. We solve this problem by proposing a circulation conflict resolving algorithm. Furthermore, video synopsis methods usually minimize a multi-objective energy function. Previous approaches optimize the multiple objectives simultaneously which needs to strike a balance between them. Instead, we propose a stepwise optimization strategy consuming less running time while producing higher quality. Experiments demonstrate the effectiveness of our method.

Object pose estimation based on a single RGB image has wide application potential but is difficult to achieve. Existing pose estimation involves various inference pipelines. One popular pipeline is to first use Convolutional Neural Networks (CNN) to predict 2D projections of 3D keypoints in a single RGB image and then calculate the 6D pose via a Perspective-n-Point (PnP) solver. Due to the gap between synthetic data and real data, the model trained on synthetic data has difficulty predicting the 6D pose accurately when applied to real data. To address the acute problem, we propose a two-stage pipeline of object pose estimation based upon multi-precision vectors and segmentation-driven (Seg-Driven) PnP. In keypoint localization stage, we first develop a CNN-based three-branch network to predict multi-precision 2D vectors pointing to 2D keypoints. Then we introduce an accurate and fast Keypoint Voting scheme of Multi-precision vectors (KVM), which computes low-precision 2D keypoints using low-precision vectors and refines 2D keypoints on mid- and high-precision vectors. In the pose calculation stage, we propose Seg-Driven PnP to refine the 3D Translation of poses and get the optimal pose by minimizing the non-overlapping area between segmented and rendered masks. The Seg-Driven PnP leverages 2D segmentation trained on real images to improve the accuracy of pose estimation trained on synthetic data, thereby reducing the synthetic-to-real gap. Extensive experiments show our approach materially outperforms state-of-the-art methods on LM and HB datasets. Importantly, our proposed method works reasonably well for weakly textured and occluded objects in diverse scenes.

Current RGBT tracking research relies on the complete multi-modality input, but modal information might miss due to some factors such as thermal sensor self-calibration and data transmission error, called modality-missing challenge in this work. To address this challenge, we propose a novel invertible prompt learning approach, which integrates the content-preserving prompts into a well-trained tracking model to adapt to various modality-missing scenarios, for robust RGBT tracking. Given one modality-missing scenario, we propose to utilize the available modality to generate the prompt of the missing modality to adapt to RGBT tracking model. However, the cross-modality gap between available and missing modalities usually causes semantic distortion and information loss in prompt generation. To handle this issue, we design the invertible prompter by incorporating the full reconstruction of the input available modality from the generated prompt. To provide a comprehensive evaluation platform, we construct several high-quality benchmark datasets, in which various modality-missing scenarios are considered to simulate real-world challenges. Extensive experiments on three modality-missing benchmark datasets show that our method achieves significant performance improvements compared with state-of-the-art methods. We have released the code and simulation datasets at: https://github.com/mmic-lcl.

While vision-language pretrained models (VLMs) excel in various multimodal understanding tasks, their potential in fine-grained audio-visual reasoning, particularly for audio-visual question answering (AVQA), remains largely unexplored. AVQA presents specific challenges for VLMs due to the requirement of visual understanding at the region level and seamless integration with audio modality. Previous VLM-based AVQA methods merely used CLIP as a feature encoder but underutilized its knowledge, and mistreated audio and video as separate entities in a dual-stream framework as most AVQA methods. This paper proposes a new CLIP-powered target-aware single-stream (TASS) network for AVQA using the pretrained knowledge of the CLIP model through the audio-visual matching characteristic of nature. It consists of two key components: the target-aware spatial grounding module (TSG+) and the single-stream joint temporal grounding module (JTG). Specifically, TSG+ module transfers the image-text matching knowledge from CLIP models to the required region-text matching process without corresponding ground-truth labels. Moreover, unlike previous separate dual-stream networks that still required an additional audio-visual fusion module, JTG unifies audio-visual fusion and question-aware temporal grounding in a simplified single-stream architecture. It treats audio and video as a cohesive entity and further extends the image-text matching knowledge to audio-text matching by preserving their temporal correlation with our proposed cross-modal synchrony (CMS) loss. Besides, we propose a simple yet effective preprocessing strategy to optimize accuracy-efficiency trade-offs. Extensive experiments conducted on the MUSIC-AVQA benchmark verified the effectiveness of our proposed method over existing state-of-the-art methods. The code is available at https://github.com/Bravo5542/CLIP-TASS.