AAAI 2024 论文列表

Image restoration aims to reconstruct a high-quality image from a degraded low-quality observation. Recently, Transformer models have achieved promising performance on image restoration tasks due to their powerful ability to model long-range dependencies. However, the quadratically growing complexity with respect to the input size makes them inapplicable to practical applications. In this paper, we develop an efficient convolutional network for image restoration by enhancing multi-scale representation learning. To this end, we propose an omni-kernel module that consists of three branches, i.e., global, large, and local branches, to learn global-to-local feature representations efficiently. Specifically, the global branch achieves a global perceptive field via the dual-domain channel attention and frequency-gated mechanism. Furthermore, to provide multi-grained receptive fields, the large branch is formulated via different shapes of depth-wise convolutions with unusually large kernel sizes. Moreover, we complement local information using a point-wise depth-wise convolution. Finally, the proposed network, dubbed OKNet, is established by inserting the omni-kernel module into the bottleneck position for efficiency. Extensive experiments demonstrate that our network achieves state-of-the-art performance on 11 benchmark datasets for three representative image restoration tasks, including image dehazing, image desnowing, and image defocus deblurring. The code is available at https://github.com/c-yn/OKNet.

Regular decision processes (RDPs) are a subclass of non-Markovian decision processes where the transition and reward functions are guarded by some regular property of the past (a lookback). While RDPs enable intuitive and succinct representation of non-Markovian decision processes, their expressive power coincides with finite-state Markov decision processes (MDPs). We introduce omega-regular decision processes (ODPs) where the non-Markovian aspect of the transition and reward functions are extended to an omega-regular lookahead over the system evolution. Semantically, these lookaheads can be considered as promises made by the decision maker or the learning agent about her future behavior. In particular, we assume that, if the promised lookaheads are not met, then the payoff to the decision maker is falsum (least desirable payoff), overriding any rewards collected by the decision maker. We enable optimization and learning for ODPs under the discounted-reward objective by reducing them to lexicographic optimization and learning over finite MDPs. We present experimental results demonstrating the effectiveness of the proposed reduction.

Video compression relies heavily on exploiting the temporal redundancy between video frames, which is usually achieved by estimating and using the motion information. The motion information is represented as optical flows in most of the existing deep video compression networks. Indeed, these networks often adopt pre-trained optical flow estimation networks for motion estimation. The optical flows, however, may be less suitable for video compression due to the following two factors. First, the optical flow estimation networks were trained to perform inter-frame prediction as accurately as possible, but the optical flows themselves may cost too many bits to encode. Second, the optical flow estimation networks were trained on synthetic data, and may not generalize well enough to real-world videos. We address the twofold limitations by enhancing the optical flows in two stages: offline and online. In the offline stage, we fine-tune a trained optical flow estimation network with the motion information provided by a traditional (non-deep) video compression scheme, e.g. H.266/VVC, as we believe the motion information of H.266/VVC achieves a better rate-distortion trade-off. In the online stage, we further optimize the latent features of the optical flows with a gradient descent-based algorithm for the video to be compressed, so as to enhance the adaptivity of the optical flows. We conduct experiments on two state-of-the-art deep video compression schemes, DCVC and DCVC-DC. Experimental results demonstrate that the proposed offline and online enhancement together achieves on average 13.4% bitrate saving for DCVC and 4.1% bitrate saving for DCVC-DC on the tested videos, without increasing the model or computational complexity of the decoder side.

Offline optimization is an emerging problem in many experimental engineering domains including protein, drug or aircraft design, where online experimentation to collect evaluation data is too expensive or dangerous. To avoid that, one has to optimize an unknown function given only its offline evaluation at a fixed set of inputs. A naive solution to this problem is to learn a surrogate model of the unknown function and optimize this surrogate instead. However, such a naive optimizer is prone to erroneous overestimation of the surrogate (possibly due to over-fitting on a biased sample of function evaluation) on inputs outside the offline dataset. Prior approaches addressing this challenge have primarily focused on learning robust surrogate models. However, their search strategies are derived from the surrogate model rather than the actual offline data. To fill this important gap, we introduce a new learning-to-search perspective for offline optimization by reformulating it as an offline reinforcement learning problem. Our proposed policy-guided gradient search approach explicitly learns the best policy for a given surrogate model created from the offline data. Our empirical results on multiple benchmarks demonstrate that the learned optimization policy can be combined with existing offline surrogates to significantly improve the optimization performance.

3D semantic occupancy has garnered considerable attention due to its abundant structural information encompassing the entire scene in autonomous driving. However, existing 3D occupancy prediction methods contend with the constraint of low-resolution 3D voxel features arising from the limitation of computational memory. To address this limitation and achieve a more fine-grained representation of 3D scenes, we propose OctOcc, a novel octree-based approach for 3D semantic occupancy prediction. OctOcc is conceptually rooted in the observation that the vast majority of 3D space is left unoccupied. Capitalizing on this insight, we endeavor to cultivate memory-efficient high-resolution 3D occupancy predictions by mitigating superfluous cross-attentions. Specifically, we devise a hierarchical octree structure that selectively generates finer-grained cross-attentions solely in potentially occupied regions. Extending our inquiry beyond 3D space, we identify analogous redundancies within another side of cross attentions, 2D images. Consequently, a 2D image feature filtering network is conceived to expunge extraneous regions. Experimental results demonstrate that the proposed OctOcc significantly outperforms existing methods on nuScenes and SemanticKITTI datasets with limited memory consumption.

While generic person re-identification has made remarkable improvement in recent years, these methods are designed under the assumption that the entire body of the person is available. This assumption brings about a significant performance degradation when suffering from occlusion caused by various obstacles in real-world applications. To address this issue, data-driven strategies have emerged to enhance the model's robustness to occlusion. Following the random erasing paradigm, these strategies typically employ randomly generated noise to supersede randomly selected image regions to simulate obstacles. However, the random strategy is not sensitive to location and content, meaning they cannot mimic real-world occlusion cases in application scenarios. To overcome this limitation and fully exploit the real scene information in datasets, this paper proposes a more intuitive and effective data-driven strategy named Saliency-Guided Patch Transfer (SPT). Combined with the vision transformer, SPT divides person instances and background obstacles using salient patch selection. By transferring person instances to different background obstacles, SPT can easily generate photo-realistic occluded samples. Furthermore, we propose an occlusion-aware Intersection over Union (OIoU) with mask-rolling to filter the more suitable combination and a class-ignoring strategy to achieve more stable processing. Extensive experimental evaluations conducted on occluded and holistic person re-identification benchmarks demonstrate that SPT provides a significant performance gain among different ViT-based ReID algorithms on occluded ReID.

Single-domain generalization (S-DG) aims to generalize a model to unseen environments with a single-source domain. However, most S-DG approaches have been conducted in the field of classification. When these approaches are applied to object detection, the semantic features of some objects can be damaged, which can lead to imprecise object localization and misclassification. To address these problems, we propose an object-aware domain generalization (OA-DG) method for single-domain generalization in object detection. Our method consists of data augmentation and training strategy, which are called OA-Mix and OA-Loss, respectively. OA-Mix generates multi-domain data with multi-level transformation and object-aware mixing strategy. OA-Loss enables models to learn domain-invariant representations for objects and backgrounds from the original and OA-Mixed images. Our proposed method outperforms state-of-the-art works on standard benchmarks. Our code is available at https://github.com/WoojuLee24/OA-DG.

This paper focuses on the Audio-Visual Question Answering (AVQA) task that aims to answer questions derived from untrimmed audible videos. To generate accurate answers, an AVQA model is expected to find the most informative audio-visual clues relevant to the given questions. In this paper, we propose to explicitly consider fine-grained visual objects in video frames (object-level clues) and explore the multi-modal relations (\textit{i.e.}, the object, audio, and question) in terms of feature interaction and model optimization. For the former, we present an end-to-end object-oriented network that adopts a question-conditioned clue discovery module to concentrate audio/visual modalities on respective keywords of the question and designs a modality-conditioned clue collection module to highlight closely associated audio segments or visual objects. For model optimization, we propose an object-aware adaptive-positivity learning strategy that selects the highly semantic-matched multi-modal pair as \textit{positivity}. Specifically, we design two object-aware contrastive loss functions to identify the highly relevant question-object pairs and audio-object pairs, respectively. These selected pairs are constrained to have larger similarity values than the mismatched pairs. The positivity-selecting process is adaptive as the positivity pairs selected in each video frame may be different. These two object-aware objectives help the model understand \textit{which objects are exactly relevant to the question} and \textit{which are making sounds}. Extensive experiments on the MUSIC-AVQA dataset demonstrate the proposed method is effective in finding favorable audio-visual clues and also achieves new state-of-the-art question-answering performance. The code is available at https://github.com/zhangbin-ai/APL.

Visual question answering is a multimodal task that requires the joint comprehension of visual and textual information. However, integrating visual and textual semantics solely through attention layers is insufficient to comprehensively understand and align information from both modalities. Intuitively, object attributes can naturally serve as a bridge to unify them, which has been overlooked in previous research. In this paper, we propose a novel VQA approach from the perspective of utilizing object attribute, aiming to achieve better object-level visual-language alignment and multimodal scene understanding. Specifically, we design an attribute fusion module and a contrastive knowledge distillation module. The attribute fusion module constructs a multimodal graph neural network to fuse attributes and visual features through message passing. The enhanced object-level visual features contribute to solving fine-grained problem like counting-question. The better object-level visual-language alignment aids in understanding multimodal scenes, thereby improving the model's robustness. Furthermore, to augment scene understanding and the out-of-distribution performance, the contrastive knowledge distillation module introduces a series of implicit knowledge. We distill knowledge into attributes through contrastive loss, which further strengthens the representation learning of attribute features and facilitates visual-linguistic alignment. Intensive experiments on six datasets, COCO-QA, VQAv2, VQA-CPv2, VQA-CPv1, VQAvs and TDIUC, show the superiority of the proposed method.

Occlusion is a common issue in 3D reconstruction from RGB-D videos, often blocking the complete reconstruction of objects and presenting an ongoing problem. In this paper, we propose a novel framework, empowered by a 2D diffusion-based in-painting model, to reconstruct complete surfaces for the hidden parts of objects. Specifically, we utilize a pre-trained diffusion model to fill in the hidden areas of 2D images. Then we use these in-painted images to optimize a neural implicit surface representation for each instance for 3D reconstruction. Since creating the in-painting masks needed for this process is tricky, we adopt a human-in-the-loop strategy that involves very little human engagement to generate high-quality masks. Moreover, some parts of objects can be totally hidden because the videos are usually shot from limited perspectives. To ensure recovering these invisible areas, we develop a cascaded network architecture for predicting signed distance field, making use of different frequency bands of positional encoding and maintaining overall smoothness. Besides the commonly used rendering loss, Eikonal loss, and silhouette loss, we adopt a CLIP-based semantic consistency loss to guide the surface from unseen camera angles. Experiments on ScanNet scenes show that our proposed framework achieves state-of-the-art accuracy and completeness in object-level reconstruction from scene-level RGB-D videos. Code: https://github.com/THU-LYJ-Lab/O2-Recon.

Large language models (LLMs) with hundreds of billions of parameters require powerful server-grade GPUs for inference, limiting their practical deployment. To address this challenge, we introduce the outlier-aware weight quantization (OWQ) method, which aims to minimize LLM's footprint through low-precision representation. OWQ prioritizes a small subset of structured weights sensitive to quantization, storing them in high-precision, while applying highly tuned quantization to the remaining dense weights. This sensitivity-aware mixed-precision scheme reduces the quantization error notably, and extensive experiments demonstrate that 3.1-bit models using OWQ perform comparably to 4-bit models optimized by OPTQ. Furthermore, OWQ incorporates a parameter-efficient fine-tuning for task-specific adaptation, called weak column tuning (WCT), enabling accurate task-specific LLM adaptation with minimal memory overhead in the optimized format. OWQ represents a notable advancement in the flexibility, efficiency, and practicality of LLM optimization literature. The source code is available at https://github.com/xvyaward/owq.

In reinforcement learning, the optimism in the face of uncertainty (OFU) is a mainstream principle for directing exploration towards less explored areas, characterized by higher uncertainty. However, in the presence of environmental stochasticity (noise), purely optimistic exploration may lead to excessive probing of high-noise areas, consequently impeding exploration efficiency. Hence, in exploring noisy environments, while optimism-driven exploration serves as a foundation, prudent attention to alleviating unnecessary over-exploration in high-noise areas becomes beneficial. In this work, we propose Optimistic Value Distribution Explorer (OVD-Explorer) to achieve a noise-aware optimistic exploration for continuous control. OVD-Explorer proposes a new measurement of the policy's exploration ability considering noise in optimistic perspectives, and leverages gradient ascent to drive exploration. Practically, OVD-Explorer can be easily integrated with continuous control RL algorithms. Extensive evaluations on the MuJoCo and GridChaos tasks demonstrate the superiority of OVD-Explorer in achieving noise-aware optimistic exploration.

Large Language Models (LLMs) have achieved human-level fluency in text generation, making it difficult to distinguish between human-written and LLM-generated texts. This poses a growing risk of misuse of LLMs and demands the development of detectors to identify LLM-generated texts. However, existing detectors lack robustness against attacks: they degrade detection accuracy by simply paraphrasing LLM-generated texts. Furthermore, a malicious user might attempt to deliberately evade the detectors based on detection results, but this has not been assumed in previous studies. In this paper, we propose OUTFOX, a framework that improves the robustness of LLM-generated-text detectors by allowing both the detector and the attacker to consider each other's output. In this framework, the attacker uses the detector's prediction labels as examples for in-context learning and adversarially generates essays that are harder to detect, while the detector uses the adversarially generated essays as examples for in-context learning to learn to detect essays from a strong attacker. Experiments in the domain of student essays show that the proposed detector improves the detection performance on the attacker-generated texts by up to +41.3 points F1-score. Furthermore, the proposed detector shows a state-of-the-art detection performance: up to 96.9 points F1-score, beating existing detectors on non-attacked texts. Finally, the proposed attacker drastically degrades the performance of detectors by up to -57.0 points F1-score, massively outperforming the baseline paraphrasing method for evading detection.

Recently, several lightweight methods have been proposed to implement single-image super-resolution (SISR) on resource-constrained devices. However, these methods primarily focus on simplifying network structures without the full utilization of shallow features. The fact remains that shallow features encompass crucial details for the super-resolution task, including edges, textures, and colors. Therefore, developing a novel architecture that can effectively integrate features from different levels and capitalize on their mutual complementarity is necessary. We first analyze the relationship between multi-stage features and the restoration tasks in a classic lightweight SR method. Based on these observations, we propose an Omni-Stage Feature Fusion (OSFF) architecture, which incorporates Original Image Stacked Initialisation, Shallow Feature Global Connection, and Multi-Receptive Field Dynamic Fusion. An Attention-Enhanced Feature Distillation module is also designed to enhance the model performance. Finally, leveraging these contributions, we construct an Omni-Stage Feature Fusion Network (OSFFNet). Through extensive experiments on various benchmark datasets, the proposed model outperforms state-of-the-art methods. Notably, it achieves a 0.26dB PSNR improvement over the second-best method for x2 SR on the Urban100 dataset.

Avoiding synthesizing specific visual concepts is an essential challenge in responsible visual synthesis. However, the visual concept that needs to be avoided for responsible visual synthesis tends to be diverse, depending on the region, context, and usage scenarios. In this work, we formalize a new task, Open-vocabulary Responsible Visual Synthesis (ORES), where the synthesis model is able to avoid forbidden visual concepts while allowing users to input any desired content. To address this problem, we present a Two-stage Intervention (TIN) framework. By introducing 1) rewriting with learnable instruction through a large-scale language model (LLM) and 2) synthesizing with prompt intervention on a diffusion synthesis model, it can effectively synthesize images avoiding any concepts but following the user's query as much as possible. To evaluate on ORES, we provide a publicly available dataset, baseline models, and benchmark. Experimental results demonstrate the effectiveness of our method in reducing risks of image generation. Our work highlights the potential of LLMs in responsible visual synthesis. Our code and dataset is public available in https://github.com/kodenii/ORES.

Online contextual reasoning and association across consecutive video frames are critical to perceive instances in visual tracking. However, most current top-performing trackers persistently lean on sparse temporal relationships between reference and search frames via an offline mode. Consequently, they can only interact independently within each image-pair and establish limited temporal correlations. To alleviate the above problem, we propose a simple, flexible and effective video-level tracking pipeline, named ODTrack, which densely associates the contextual relationships of video frames in an online token propagation manner. ODTrack receives video frames of arbitrary length to capture the spatio-temporal trajectory relationships of an instance, and compresses the discrimination features (localization information) of a target into a token sequence to achieve frame-to-frame association. This new solution brings the following benefits: 1) the purified token sequences can serve as prompts for the inference in the next video frame, whereby past information is leveraged to guide future inference; 2) the complex online update strategies are effectively avoided by the iterative propagation of token sequences, and thus we can achieve more efficient model representation and computation. ODTrack achieves a new SOTA performance on seven benchmarks, while running at real-time speed. Code and models are available at https://github.com/GXNU-ZhongLab/ODTrack.

Model-based offline reinforcement learning (RL) algorithms have emerged as a promising paradigm for offline RL. These algorithms usually learn a dynamics model from a static dataset of transitions, use the model to generate synthetic trajectories, and perform conservative policy optimization within these trajectories. However, our observations indicate that policy optimization methods used in these model-based offline RL algorithms are not effective at exploring the learned model and induce biased exploration, which ultimately impairs the performance of the algorithm. To address this issue, we propose Offline Conservative ExplorAtioN (OCEAN), a novel rollout approach to model-based offline RL. In our method, we incorporate additional exploration techniques and introduce three conservative constraints based on uncertainty estimation to mitigate the potential impact of significant dynamic errors resulting from exploratory transitions. Our work is a plug-in method and can be combined with classical model-based RL algorithms, such as MOPO, COMBO, and RAMBO. Experiment results of our method on the D4RL MuJoCo benchmark show that OCEAN significantly improves the performance of existing algorithms.

Consistency and interpretability have long been the critical issues in MRI reconstruction. While interpretability has been dramatically improved with the employment of deep unfolding networks (DUNs), current methods still suffer from inconsistencies and generate inferior anatomical structure. Especially in multi-contrast scenes, different imaging protocols often exacerbate the concerned issue. In this paper, we propose a range-null decomposition-assisted DUN architecture to ensure consistency while still providing desirable interpretability. Given the input decomposed, we argue that the inconsistency could be analytically relieved by feeding solely the null-space component into proximal mapping, while leaving the range-space counterpart fixed. More importantly, a correlation decoupling scheme is further proposed to narrow the information gap for multi-contrast fusion, which dynamically borrows isotropic features from the opponent while maintaining the modality-specific ones. Specifically, the two features are attached to different frequencies and learned individually by the newly designed isotropy encoder and anisotropy encoder. The former strives for the contrast-shared information, while the latter serves to capture the contrast-specific features. The quantitative and qualitative results show that our proposal outperforms most cutting-edge methods by a large margin. Codes will be released on https://github.com/chenjiachengzzz/RNU.

We introduce a novel visual question answering (VQA) task in the context of autonomous driving, aiming to answer natural language questions based on street-view clues. Compared to traditional VQA tasks, VQA in autonomous driving scenario presents more challenges. Firstly, the raw visual data are multi-modal, including images and point clouds captured by camera and LiDAR, respectively. Secondly, the data are multi-frame due to the continuous, real-time acquisition. Thirdly, the outdoor scenes exhibit both moving foreground and static background. Existing VQA benchmarks fail to adequately address these complexities. To bridge this gap, we propose NuScenes-QA, the first benchmark for VQA in the autonomous driving scenario, encompassing 34K visual scenes and 460K question-answer pairs. Specifically, we leverage existing 3D detection annotations to generate scene graphs and design question templates manually. Subsequently, the question-answer pairs are generated programmatically based on these templates. Comprehensive statistics prove that our NuScenes-QA is a balanced large-scale benchmark with diverse question formats. Built upon it, we develop a series of baselines that employ advanced 3D detection and VQA techniques. Our extensive experiments highlight the challenges posed by this new task. Codes and dataset are available at https://github.com/qiantianwen/NuScenes-QA.