ICCV 2023 Papers

Estimating the transferability of publicly available pre-trained models to a target task has assumed an important place for transfer learning tasks in recent years. Existing efforts propose metrics that allow a user to choose one model from a pool of pre-trained models without having to fine-tune each model individually and identify one explicitly. With the growth in the number of available pre-trained models and the popularity of model ensembles, it also becomes essential to study the transferability of multiple-source models for a given target task. The few existing efforts study transferability in such multi-source ensemble settings using just the outputs of the classification layer and neglect possible domain or task mismatch. Moreover, they overlook the most important factor while selecting the source models, viz., the cohesiveness factor between them, which can impact the performance and confidence in the prediction of the ensemble. To address these gaps, we propose a novel Optimal tranSport-based suBmOdular tRaNsferability metric (OSBORN) to estimate the transferability of an ensemble of models to a downstream task. OSBORN collectively accounts for image domain difference, task difference, and cohesiveness of models in the ensemble to provide reliable estimates of transferability. We gauge the performance of OSBORN on both image classification and semantic segmentation tasks. Our setup includes 28 source datasets, 11 target datasets, 5 model architectures, and 2 pre-training methods. We benchmark our method against current state-of-the-art metrics MS-LEEP and E-LEEP, and outperform them consistently using the proposed approach.

Thanks to the excellent global modeling capability of attention mechanisms, the Vision Transformer has achieved better results than ConvNet in many computer tasks. However, in generating hierarchical feature maps, the Transformer still adopts the ConvNet feature aggregation scheme. This leads to the problem that the semantic information of the grid area of image becomes confused after feature aggregation, making it difficult for attention to accurately model global relationships. To address this, we propose the Hierarchy Aware Feature Aggregation framework (HAFA). HAFA enhances the extraction of local features adaptively in shallow layers where semantic information is weak, while is able to aggregate patches with similar semantics in deep layers. The clear semantic information of the aggregated patches, enables the attention mechanism to more accurately model global information at the semantic level. Extensive experiments show that after using the HAFA framework, significant improvements have been achieved relative to the baseline models in image classification, object detection, and semantic segmentation tasks.

In this paper, we focus on the mural-restoration task, which aims to detect damaged regions in the mural and repaint them automatically. Different from traditional image restoration tasks like in/out/blind-painting and image renovation, the corrupted mural suffers from more complicated degradation. However, existing mural-restoration methods and datasets still focus on simple degradation like masking. Such a significant gap prevents mural-restoration from being applied to real scenarios. To fill this gap, in this work, we propose a systematic framework to simulate the physical process for damaged murals and provide a new benchmark dataset for mural-restoration. Limited by the simplification of the data synthesis process, the previous mural-restoration methods suffer from poor performance in our proposed dataset. To handle this problem, we propose the Attention Diffusion Framework (ADF) for this challenging task. Within the framework, a damage attention map module is proposed to estimate the damage extent. Facing the diversity of defects, we propose a series of loss functions to choose repair strategies adaptively. Finally, experimental results support the effectiveness of the proposed framework in terms of both mural synthesis and restoration.

The development of face editing has been boosted since the birth of StyleGAN. While previous works have explored different interactive methods, such as sketching and exemplar photos, they have been limited in terms of expressiveness and generality. In this paper, we propose a new interaction method by guiding the editing with abstract clipart, composed of a set of simple semantic parts, allowing users to control across face photos with simple clicks. However, this is a challenging task given the large domain gap between colorful face photos and abstract clipart with limited data. To solve this problem, we introduce a framework called ClipFaceShop built on top of StyleGAN. The key idea is to take advantage of W+ latent code encoded rich and disentangled visual features, and create a new lightweight selective feature adaptor to predict a modifiable path toward the target output photo. Since no pairwise labeled data exists for training, we design a set of losses to provide supervision signals for learning the modifiable path. Experimental results show that ClipFaceShop generates realistic and faithful face photos, sharing the same facial attributes as the reference clipart. We demonstrate that ClipFaceShop supports clipart in diverse styles, even in form of a free-hand sketch.

Parameter efficient tuning (PET) has received considerable attention owing to its applicability to reduce the number of parameters that need to be updated while maintaining competitive performance and providing better hardware resource savings. Although substantial progress has been made, most existing studies mainly focus on either single-modal tasks or simple classification tasks, with few works paying attention to the dense prediction tasks and the interaction between different modalities. Therefore, in this paper, we do an in-depth investigation of the efficient tuning problem on referring image segmentation. First, considering the absence of interaction between the dual encoder, we design a novel adapter named Bridger to facilitate the exchange of cross-modal information. This module also plays a role in injecting vision-specific inductive biases and task-specific information into the pre-trained model while keeping its original parameters fixed. Second, we design a lightweight decoder for referring image segmentation to make further alignment on visual and linguistic features. To perform a comprehensive assessment and promote further research, we evaluate the proposed framework on several challenging benchmarks. Experimental results illustrate the effectiveness of our approach. Updating only 1.61% to 3.38% parameters, the proposed framework gains comparable or even superior performance compared to existing full fine-tuning methods that utilize the same backbone.

Knowledge distillation (KD) has shown potential for learning compact models in dense object detection. However, the commonly used softmax-based distillation ignores the absolute classification scores for individual categories. Thus, the optimum of the distillation loss does not necessarily lead to the optimal student classification scores for dense object detectors. This cross-task protocol inconsistency is critical, especially for dense object detectors, since the foreground categories are extremely imbalanced. To address the issue of protocol differences between distillation and classification, we propose a novel distillation method with cross-task consistent protocols, tailored for the dense object detection. For classification distillation, we address the cross-task protocol inconsistency problem by formulating the classification logit maps in both teacher and student models as multiple binary-classification maps and applying a binary-classification distillation loss to each map. For localization distillation, we design an IoU-based Localization Distillation Loss that is free from specific network structures and can be compared with existing localization distillation losses. Our proposed method is simple but effective, and experimental results demonstrate its superiority over existing methods.

Given its wide applications, there is increasing focus on generating 3D human motions from textual descriptions. Differing from the majority of previous works, which regard actions as single entities and can only generate short sequences for simple motions, we propose EMS, an elaborative motion synthesis model conditioned on detailed natural language descriptions. It generates natural and smooth motion sequences for long and complicated actions by factorizing them into groups of atomic actions. Meanwhile, it understands atomic-action level attributes (e.g., motion direction, speed, and body parts) and enables users to generate sequences of unseen complex actions from unique sequences of known atomic actions with independent attribute settings and timings applied. We evaluate our method on the KIT Motion-Language and BABEL benchmarks, where it outperforms all previous state-of-the-art with noticeable margins.

As video analysis using deep learning models becomes more widespread, the vulnerability of such models to adversarial attacks is becoming a pressing concern. In particular, Universal Adversarial Perturbation (UAP) poses a significant threat, as a single perturbation can mislead deep learning models on entire datasets. We propose a novel video UAP using image data and image model. This enables us to take advantage of the rich image data and image model-based studies available for video applications. However, there is a challenge that image models are limited in their ability to analyze the temporal aspects of videos, which is crucial for a successful video attack. To address this challenge, we introduce the Breaking Temporal Consistancy (BTC) method, which is the first attempt to incorporate temporal information into video attacks using image models. We aim to generate adversarial videos that have opposite patterns to the original. Specifically, BTC-UAP minimizes the feature similarity between neighboring frames in videos. Our approach is simple but effective at attacking unseen video models. Additionally, it is applicable to videos of varying lengths and invariant to temporal shifts. Our approach surpasses existing methods in terms of effectiveness on various datasets, including ImageNet, UCF-101, and Kinetics-400.

Weird, unusual, and uncanny images pique the curiosity of observers because they challenge commonsense. For example, an image released during the 2022 world cup depicts the famous soccer stars Lionel Messi and Cristiano Ronaldo playing chess, which playfully violates our expectation that their competition should occur on the football field. Humans can easily recognize and interpret these unconventional images, but can AI models do the same? We introduce WHOOPS!, a new dataset and benchmark for visual commonsense. The dataset is comprised of purposefully commonsense-defying images created by designers using publicly-available image generation tools like Midjourney. We consider several tasks posed over the dataset. In addition to image captioning, cross-modal matching, and visual question answering, we introduce a difficult explanation generation task, where models must identify and explain why a given image is unusual. Our results show that state-of-the-art models such as GPT3 and BLIP2 still lag behind human performance on WHOOPS!. We hope our dataset will inspire the development of AI models with stronger visual commonsense reasoning abilities.

Box-supervised instance segmentation has gained much attention as it requires only simple box annotations instead of costly mask or polygon annotations. However, existing box-supervised instance segmentation models mainly focus on mask-based frameworks. We propose a new end-to-end training technique, termed BoxSnake, to achieve effective polygonal instance segmentation using only box annotations for the first time. Our method consists of two loss functions: (1) a point-based unary loss that constrains the bounding box of predicted polygons to achieve coarse-grained segmentation; and (2) a distance-aware pairwise loss that encourages the predicted polygons to fit the object boundaries. Compared with the mask-based weakly-supervised methods, BoxSnake further reduces the performance gap between the predicted segmentation and the bounding box, and shows significant superiority on the Cityscapes dataset.

Recent text-to-image diffusion models have demonstrated an astonishing capacity to generate high-quality images. However, researchers mainly studied the way of synthesizing images with only text prompts. While some works have explored using other modalities as conditions, considerable paired data, e.g., box/mask-image pairs, and fine-tuning time are required for nurturing models. As such paired data is time-consuming and labor-intensive to acquire and restricted to a closed set, this potentially becomes the bottleneck for applications in an open world. This paper focuses on the simplest form of user-provided conditions, e.g., box or scribble. To mitigate the aforementioned problem, we propose a training-free method to control objects and contexts in the synthesized images adhering to the given spatial conditions. Specifically, three spatial constraints, i.e., Inner-Box, Outer-Box, and Corner Constraints, are designed and seamlessly integrated into the denoising step of diffusion models, requiring no additional training and massive annotated layout data. Extensive experimental results demonstrate that the proposed constraints can control what and where to present in the images while retaining the ability of Diffusion models to synthesize with high fidelity and diverse concept coverage. The code is publicly available at https://github.com/showlab/BoxDiff.

It has been established that training a box-based detector network can enhance the localization performance of weakly supervised and unsupervised methods. Moreover, we extend this understanding by demonstrating that these detectors can be utilized to improve the original network, paving the way for further advancements. To accomplish this, we train the detectors on top of the network output instead of the image data and apply suitable loss backpropagation. Our findings reveal a significant improvement in phrase grounding for the "what is where by looking" task, as well as various methods of unsupervised object discovery.

Recent deep learning methods have achieved superior results in shadow removal. However, most of these supervised methods rely on training over a huge amount of shadow and shadow-free image pairs, which require laborious annotations and may end up with poor model generalization. Shadows, in fact, only form partial degradation in images, while their non-shadow regions provide rich structural information potentially for unsupervised learning. In this paper, we propose a novel diffusion-based solution for unsupervised shadow removal, which separately models the shadow, non-shadow, and their boundary regions. We employ a pretrained unconditional diffusion model fused with non-corrupted information to generate the natural shadow-free image. While the diffusion model can restore the clear structure in the boundary region by utilizing its adjacent non-corrupted contextual information, it fails to address the inner shadow area due to the isolation of the non-corrupted contexts. Thus we further propose a Shadow-Invariant Intrinsic Decomposition module to exploit the underlying reflectance in the shadow region to maintain structural consistency during the diffusive sampling. Extensive experiments on the publicly available shadow removal datasets show that the proposed method achieves a significant improvement compared to existing unsupervised methods, and even is comparable with some existing supervised methods.

Although considerable progress has been made in the deraining task under synthetic data, it is still a tough problem under real rain scenes, due to the domain gap between the synthetic and real data. Besides, difficulties in collecting and labeling diverse real rain images hinder the progress of this field. Consequently, we attempt to promote real rain removal from rain image generation (RIG) perspective. Existing RIG methods mainly focus on diversity but miss realistic, or the realistic but neglect diversity of the generation. To solve this dilemma, we propose a physical alignment and controllable generation network (PCGNet) for diverse and realistic rain generation. Our key idea is to simultaneously utilize the controllability of attributes from synthetic and the realism of appearance from real data. Specifically, we devise a unified framework to disentangle background, rain attributes, and appearance style from synthetic and real data. Then we collaboratively align the factors with a novel semi-supervised weight moving strategy for attribute, an explicit distribution modeling method for real rain style. Furthermore, we pack these aligned factors into the generation model, achieving physical controllable mapping from the attributes to real rainy with image-level and attribute-level consistency loss. Extensive experiments show that PCGNet can effectively generate appealing rainy results, which sifnicantltly improve the performance under synthetic and real scenes for all existing deraining methods.

The development of vision models for real-world applications is hindered by the challenge of annotated data scarcity, which has necessitated the adoption of data-efficient visual learning techniques such as semi-supervised learning. Unfortunately, the prevalent cross-entropy supervision is limited by its focus on category discrimination while disregarding the semantic connection between concepts, which ultimately results in the suboptimal exploitation of scarce labeled data. To address this issue, this paper presents a novel approach that seeks to leverage linguistic knowledge for data-efficient visual learning. The proposed approac, BorLan, Borrows knowledge from off-the-shelf pretrained Language models that are already endowed with rich semantics extracted from large corpora, to compensate the semantic deficiency due to limited annotation in visual training. Specifically, we design a distribution alignment objective, which guides the vision model to learn both semantic-aware and domain-agnostic representations for the task through linguistic knowledge. One significant advantage of this paradigm is its flexibility in combining various visual and linguistic models. Extensive experiments on semi-supervised learning, single domain generalization and few-shot learning validate its effectiveness.

We present a novel framework to bootstrap Motion forecasting with Self-consistent Constraints (MISC). The motion forecasting task aims at predicting future trajectories of vehicles by incorporating spatial and temporal information from the past. A key design of MISC is the proposed Dual Consistency Constraints that regularize the predicted trajectories under spatial and temporal perturbation during training. Also, to model the multi-modality in motion forecasting, we design a novel self-ensembling scheme to obtain accurate teacher targets to enforce the self-constraints with multi-modality supervision. With explicit constraints from multiple teacher targets, we observe a clear improvement in the prediction performance. Extensive experiments on the Argoverse motion forecasting benchmark and Waymo Open Motion dataset show that MISC significantly outperforms the state-of-the-art methods. As the proposed strategies are general and can be easily incorporated into other motion forecasting approaches, we also demonstrate that our proposed scheme consistently improves the prediction performance of several existing methods.

Bag-based multiple instance learning (MIL) methods have become the mainstream for Whole Slide Image (WSI) classification. However, there are still three important issues that have not been fully addressed: (1) positive bags with a low positive instance ratio are prone to the influence of a large number of negative instances; (2) the correlation between local and global features of pathology images has not been fully modeled; and (3) there is a lack of effective information interaction between different magnifications. In this paper, we propose MILBooster, a powerful dual-scale multi-stage MIL framework to address these issues from the perspectives of distribution, correlation, and magnification. Specifically, to address issue (1), we propose a plug-and-play bag filter that effectively increases the positive instance ratio of positive bags. For issue (2), we propose a novel window-based Transformer architecture called PiceBlock to model the correlation between local and global features of pathology images. For issue (3), we propose a dual-branch architecture to process different magnifications and design an information interaction module called Scale Mixer for efficient information interaction between them. We conducted extensive experiments on four clinical WSI classification tasks using three datasets. MILBooster achieved new state-of-the-art performance on all these tasks. Codes will be available.

Although deep learning has significantly facilitated the progress of single image super-resolution (SISR) in recent years, it still hits bottlenecks to further improve SR performance with the continuous growth of model scale. Therefore, one of the hotspots in the field is to construct efficient SISR models by elevating the effectiveness of feature representation. In this work, we present a straightforward and generic approach for feature enhancement that can effectively promote the performance of SR models, dubbed partial channel shifting (PCS). Specifically, it is inspired by the temporal shifting in video understanding and displaces part of the channels along the spatial dimensions, thus allowing the effective receptive field to be amplified and the feature diversity to be augmented at almost zero cost. Also, it can be assembled into off-the-shelf models as a plug-and-play component for performance boosting without extra network parameters and computational overhead. However, regulating the features with PCS encounters some issues, like shifting directions and amplitudes, proportions, and patterns of shifted channels, etc. We impose some technical constraints on the issues to simplify the general channel shifting. Extensive and throughout experiments illustrate that the PCS indeed enlarges the effective receptive field, augments the feature diversity for efficiently enhancing SR recovery, and can endow obvious performance gains to existing models.

Semantic segmentation is a computer vision task that associates a label with each pixel in an image. Modern approaches tend to introduce class embeddings into semantic segmentation for deeply utilizing category semantics, and regard supervised class masks as final predictions. In this paper, we explore the mechanism of class embeddings and have an insight that more explicit and meaningful class embeddings can be generated based on class masks purposely. Following this observation, we propose ECENet, a new segmentation paradigm, in which class embeddings are obtained and enhanced explicitly during interacting with multi-stage image features. Based on this, we revisit the traditional decoding process and explore inverted information flow between segmentation masks and class embeddings. Furthermore, to ensure the discriminability and informativity of features from backbone, we propose a Feature Reconstruction module, which combines intrinsic and diverse branches together to ensure the concurrence of diversity and redundancy in features. Experiments show that our ECENet outperforms its counterparts on the ADE20K dataset with much less computational cost and achieves new state-of-the-art results on PASCAL-Context dataset. The code will be released at https://gitee.com/mindspore/models and https://github.com/Carol-lyh/ECENet.