Zhenyu Zhang

Zhiqiang Yan, Kun Wang, Xiang Li et al.

Unsupervised depth completion aims to recover dense depth from the sparse one without using the ground-truth annotation. Although depth measurement obtained from LiDAR is usually sparse, it contains valid and real distance information, i.e., scale-consistent absolute depth values. Meanwhile, scale-agnostic counterparts seek to estimate relative depth and have achieved impressive performance. To leverage both the inherent characteristics, we thus suggest to model scale-consistent depth upon unsupervised scale-agnostic frameworks. Specifically, we propose the decomposed scale-consistent learning (DSCL) strategy, which disintegrates the absolute depth into relative depth prediction and global scale estimation, contributing to individual learning benefits. But unfortunately, most existing unsupervised scale-agnostic frameworks heavily suffer from depth holes due to the extremely sparse depth input and weak supervised signal. To tackle this issue, we introduce the global depth guidance (GDG) module, which attentively propagates dense depth reference into the sparse target via novel dense-to-sparse attention. Extensive experiments show the superiority of our method on outdoor KITTI benchmark, ranking 1st and outperforming the best KBNet more than 12% in RMSE. In addition, our approach achieves state-of-the-art performance on indoor NYUv2 dataset.

Zhida Feng, Zhenyu Zhang, Xintong Yu et al.

Recent progress in diffusion models has revolutionized the popular technology of text-to-image generation. While existing approaches could produce photorealistic high-resolution images with text conditions, there are still several open problems to be solved, which limits the further improvement of image fidelity and text relevancy. In this paper, we propose ERNIE-ViLG 2.0, a large-scale Chinese text-to-image diffusion model, to progressively upgrade the quality of generated images by: (1) incorporating fine-grained textual and visual knowledge of key elements in the scene, and (2) utilizing different denoising experts at different denoising stages. With the proposed mechanisms, ERNIE-ViLG 2.0 not only achieves a new state-of-the-art on MS-COCO with zero-shot FID score of 6.75, but also significantly outperforms recent models in terms of image fidelity and image-text alignment, with side-by-side human evaluation on the bilingual prompt set ViLG-300.

Kun Wang, Zhiqiang Yan, Zhenyu Zhang et al.

Reconstructing category-specific objects using Neural Radiance Field (NeRF) from a single image is a promising yet challenging task. Existing approaches predominantly rely on projection-based feature retrieval to associate 3D points in the radiance field with local image features from the reference image. However, this process is computationally expensive, dependent on known camera intrinsics, and susceptible to occlusions. To address these limitations, we propose Variable Radiance Field (VRF), a novel framework capable of efficiently reconstructing category-specific objects without requiring known camera intrinsics and demonstrating robustness against occlusions. First, we replace the local feature retrieval with global latent representations, generated through a single feed-forward pass, which improves efficiency and eliminates reliance on camera intrinsics. Second, to tackle coordinate inconsistencies inherent in real-world dataset, we define a canonical space by introducing a learnable, category-specific shape template and explicitly aligning each training object to this template using a learnable 3D transformation. This approach also reduces the complexity of geometry prediction to modeling deformations from the template to individual instances. Finally, we employ a hyper-network-based method for efficient NeRF creation and enhance the reconstruction performance through a contrastive learning-based pretraining strategy. Evaluations on the CO3D dataset demonstrate that VRF achieves state-of-the-art performance in both reconstruction quality and computational efficiency.

Zhenyu Zhang, Bowen Yu, Haiyang Yu et al.

Building document-grounded dialogue systems have received growing interest as documents convey a wealth of human knowledge and commonly exist in enterprises. Wherein, how to comprehend and retrieve information from documents is a challenging research problem. Previous work ignores the visual property of documents and treats them as plain text, resulting in incomplete modality. In this paper, we propose a Layout-aware document-level Information Extraction dataset, LIE, to facilitate the study of extracting both structural and semantic knowledge from visually rich documents (VRDs), so as to generate accurate responses in dialogue systems. LIE contains 62k annotations of three extraction tasks from 4,061 pages in product and official documents, becoming the largest VRD-based information extraction dataset to the best of our knowledge. We also develop benchmark methods that extend the token-based language model to consider layout features like humans. Empirical results show that layout is critical for VRD-based extraction, and system demonstration also verifies that the extracted knowledge can help locate the answers that users care about.

Shiyao Cui, Zhenyu Zhang, Yilong Chen et al.

The widespread of generative artificial intelligence has heightened concerns about the potential harms posed by AI-generated texts, primarily stemming from factoid, unfair, and toxic content. Previous researchers have invested much effort in assessing the harmlessness of generative language models. However, existing benchmarks are struggling in the era of large language models (LLMs), due to the stronger language generation and instruction following capabilities, as well as wider applications. In this paper, we propose FFT, a new benchmark with 2116 elaborated-designed instances, for LLM harmlessness evaluation with factuality, fairness, and toxicity. To investigate the potential harms of LLMs, we evaluate 9 representative LLMs covering various parameter scales, training stages, and creators. Experiments show that the harmlessness of LLMs is still under-satisfactory, and extensive analysis derives some insightful findings that could inspire future research for harmless LLM research.

Yuchen Feng, Zhenyu Zhang, Naibin Gu et al.

Multimodal large language models (MLLMs) have achieved remarkable progress on various vision-language tasks, yet their visual perception remains limited. Humans, in comparison, perceive complex scenes efficiently by dynamically scanning and focusing on salient regions in a sequential "blink-like" process. Motivated by this strategy, we first investigate whether MLLMs exhibit similar behavior. Our pilot analysis reveals that MLLMs naturally attend to different visual regions across layers and that selectively allocating more computation to salient tokens can enhance visual perception. Building on this insight, we propose Blink, a dynamic visual token resolution framework that emulates the human-inspired process within a single forward pass. Specifically, Blink includes two modules: saliency-guided scanning and dynamic token resolution. It first estimates the saliency of visual tokens in each layer based on the attention map, and extends important tokens through a plug-and-play token super-resolution (TokenSR) module. In the next layer, it drops the extended tokens when they lose focus. This dynamic mechanism balances broad exploration and fine-grained focus, thereby enhancing visual perception adaptively and efficiently. Extensive experiments validate Blink, demonstrating its effectiveness in enhancing visual perception and multimodal understanding.

Zihang Liu, Zhenyu Zhang, Hao Tang

Diffusion-based image super-resolution (SR) methods have demonstrated remarkable performance. Recent advancements have introduced deterministic sampling processes that reduce inference from 15 iterative steps to a single step, thereby significantly improving the inference speed of existing diffusion models. However, their efficiency remains limited when handling complex semantic regions due to the single-step inference. To address this limitation, we propose SAMSR, a semantic-guided diffusion framework that incorporates semantic segmentation masks into the sampling process. Specifically, we introduce the SAM-Noise Module, which refines Gaussian noise using segmentation masks to preserve spatial and semantic features. Furthermore, we develop a pixel-wise sampling strategy that dynamically adjusts the residual transfer rate and noise strength based on pixel-level semantic weights, prioritizing semantically rich regions during the diffusion process. To enhance model training, we also propose a semantic consistency loss, which aligns pixel-wise semantic weights between predictions and ground truth. Extensive experiments on both real-world and synthetic datasets demonstrate that SAMSR significantly improves perceptual quality and detail recovery, particularly in semantically complex images. Our code is released at https://github.com/Liu-Zihang/SAMSR.

Zhiqiang Yan, Yuankai Lin, Kun Wang et al.

Depth completion is a vital task for autonomous driving, as it involves reconstructing the precise 3D geometry of a scene from sparse and noisy depth measurements. However, most existing methods either rely only on 2D depth representations or directly incorporate raw 3D point clouds for compensation, which are still insufficient to capture the fine-grained 3D geometry of the scene. To address this challenge, we introduce Tri-Perspective view Decomposition (TPVD), a novel framework that can explicitly model 3D geometry. In particular, (1) TPVD ingeniously decomposes the original point cloud into three 2D views, one of which corresponds to the sparse depth input. (2) We design TPV Fusion to update the 2D TPV features through recurrent 2D-3D-2D aggregation, where a Distance-Aware Spherical Convolution (DASC) is applied. (3) By adaptively choosing TPV affinitive neighbors, the newly proposed Geometric Spatial Propagation Network (GSPN) further improves the geometric consistency. As a result, our TPVD outperforms existing methods on KITTI, NYUv2, and SUN RGBD. Furthermore, we build a novel depth completion dataset named TOFDC, which is acquired by the time-of-flight (TOF) sensor and the color camera on smartphones. Project page: https://yanzq95.github.io/projectpage/TOFDC/index.html

Rui Xie, Chen Zhao, Kai Zhang et al.

Blind super-resolution methods based on stable diffusion showcase formidable generative capabilities in reconstructing clear high-resolution images with intricate details from low-resolution inputs. However, their practical applicability is often hampered by poor efficiency, stemming from the requirement of thousands or hundreds of sampling steps. Inspired by the efficient adversarial diffusion distillation (ADD), we design~\name~to address this issue by incorporating the ideas of both distillation and ControlNet. Specifically, we first propose a prediction-based self-refinement strategy to provide high-frequency information in the student model output with marginal additional time cost. Furthermore, we refine the training process by employing HR images, rather than LR images, to regulate the teacher model, providing a more robust constraint for distillation. Second, we introduce a timestep-adaptive ADD to address the perception-distortion imbalance problem introduced by original ADD. Extensive experiments demonstrate our~\name~generates better restoration results, while achieving faster speed than previous SD-based state-of-the-art models (e.g., $7$$\times$ faster than SeeSR).

Rui Xie, Yinhong Liu, Penghao Zhou et al.

Image diffusion models have been adapted for real-world video super-resolution to tackle over-smoothing issues in GAN-based methods. However, these models struggle to maintain temporal consistency, as they are trained on static images, limiting their ability to capture temporal dynamics effectively. Integrating text-to-video (T2V) models into video super-resolution for improved temporal modeling is straightforward. However, two key challenges remain: artifacts introduced by complex degradations in real-world scenarios, and compromised fidelity due to the strong generative capacity of powerful T2V models (\textit{e.g.}, CogVideoX-5B). To enhance the spatio-temporal quality of restored videos, we introduce\textbf{~\name} (\textbf{S}patial-\textbf{T}emporal \textbf{A}ugmentation with T2V models for \textbf{R}eal-world video super-resolution), a novel approach that leverages T2V models for real-world video super-resolution, achieving realistic spatial details and robust temporal consistency. Specifically, we introduce a Local Information Enhancement Module (LIEM) before the global attention block to enrich local details and mitigate degradation artifacts. Moreover, we propose a Dynamic Frequency (DF) Loss to reinforce fidelity, guiding the model to focus on different frequency components across diffusion steps. Extensive experiments demonstrate\textbf{~\name}~outperforms state-of-the-art methods on both synthetic and real-world datasets.

Haijie Yang, Zhenyu Zhang, Hao Tang et al.

Diffusion models have shown impressive potential on talking head generation. While plausible appearance and talking effect are achieved, these methods still suffer from temporal, 3D or expression inconsistency due to the error accumulation and inherent limitation of single-image generation ability. In this paper, we propose ConsistentAvatar, a novel framework for fully consistent and high-fidelity talking avatar generation. Instead of directly employing multi-modal conditions to the diffusion process, our method learns to first model the temporal representation for stability between adjacent frames. Specifically, we propose a Temporally-Sensitive Detail (TSD) map containing high-frequency feature and contours that vary significantly along the time axis. Using a temporal consistent diffusion module, we learn to align TSD of the initial result to that of the video frame ground truth. The final avatar is generated by a fully consistent diffusion module, conditioned on the aligned TSD, rough head normal, and emotion prompt embedding. We find that the aligned TSD, which represents the temporal patterns, constrains the diffusion process to generate temporally stable talking head. Further, its reliable guidance complements the inaccuracy of other conditions, suppressing the accumulated error while improving the consistency on various aspects. Extensive experiments demonstrate that ConsistentAvatar outperforms the state-of-the-art methods on the generated appearance, 3D, expression and temporal consistency. Project page: https://njust-yang.github.io/ConsistentAvatar.github.io/

Hao Yang, Zhenyu Zhang, Yanyan Zhao et al.

As a fine-grained task, multimodal aspect-based sentiment analysis (MABSA) mainly focuses on identifying aspect-level sentiment information in the text-image pair. However, we observe that it is difficult to recognize the sentiment of aspects in low-quality samples, such as those with low-resolution images that tend to contain noise. And in the real world, the quality of data usually varies for different samples, such noise is called data uncertainty. But previous works for the MABSA task treat different quality samples with the same importance and ignored the influence of data uncertainty. In this paper, we propose a novel data uncertainty-aware multimodal aspect-based sentiment analysis approach, UA-MABSA, which weighted the loss of different samples by the data quality and difficulty. UA-MABSA adopts a novel quality assessment strategy that takes into account both the image quality and the aspect-based cross-modal relevance, thus enabling the model to pay more attention to high-quality and challenging samples. Extensive experiments show that our method achieves state-of-the-art (SOTA) performance on the Twitter-2015 dataset. Further analysis demonstrates the effectiveness of the quality assessment strategy.

Zhiqiang Yan, Xiang Li, Le Hui et al.

Depth completion aims to recover dense depth maps from sparse ones, where color images are often used to facilitate this task. Recent depth methods primarily focus on image guided learning frameworks. However, blurry guidance in the image and unclear structure in the depth still impede their performance. To tackle these challenges, we explore a repetitive design in our image guided network to gradually and sufficiently recover depth values. Specifically, the repetition is embodied in both the image guidance branch and depth generation branch. In the former branch, we design a dense repetitive hourglass network (DRHN) to extract discriminative image features of complex environments, which can provide powerful contextual instruction for depth prediction. In the latter branch, we present a repetitive guidance (RG) module based on dynamic convolution, in which an efficient convolution factorization is proposed to reduce the complexity while modeling high-frequency structures progressively. Furthermore, in the semantic guidance branch, we utilize the well-known large vision model, i.e., segment anything (SAM), to supply RG with semantic prior. In addition, we propose a region-aware spatial propagation network (RASPN) for further depth refinement based on the semantic prior constraint. Finally, we collect a new dataset termed TOFDC for the depth completion task, which is acquired by the time-of-flight (TOF) sensor and the color camera on smartphones. Extensive experiments demonstrate that our method achieves state-of-the-art performance on KITTI, NYUv2, Matterport3D, 3D60, VKITTI, and our TOFDC.

Kun Wang, Zhenyu Zhang, Zhiqiang Yan et al.

Monocular depth estimation aims at predicting depth from a single image or video. Recently, self-supervised methods draw much attention since they are free of depth annotations and achieve impressive performance on several daytime benchmarks. However, they produce weird outputs in more challenging nighttime scenarios because of low visibility and varying illuminations, which bring weak textures and break brightness-consistency assumption, respectively. To address these problems, in this paper we propose a novel framework with several improvements: (1) we introduce Priors-Based Regularization to learn distribution knowledge from unpaired depth maps and prevent model from being incorrectly trained; (2) we leverage Mapping-Consistent Image Enhancement module to enhance image visibility and contrast while maintaining brightness consistency; and (3) we present Statistics-Based Mask strategy to tune the number of removed pixels within textureless regions, using dynamic statistics. Experimental results demonstrate the effectiveness of each component. Meanwhile, our framework achieves remarkable improvements and state-of-the-art results on two nighttime datasets.

Zhiqiang Yan, Kun Wang, Xiang Li et al.

Depth completion deals with the problem of recovering dense depth maps from sparse ones, where color images are often used to facilitate this task. Recent approaches mainly focus on image guided learning frameworks to predict dense depth. However, blurry guidance in the image and unclear structure in the depth still impede the performance of the image guided frameworks. To tackle these problems, we explore a repetitive design in our image guided network to gradually and sufficiently recover depth values. Specifically, the repetition is embodied in both the image guidance branch and depth generation branch. In the former branch, we design a repetitive hourglass network to extract discriminative image features of complex environments, which can provide powerful contextual instruction for depth prediction. In the latter branch, we introduce a repetitive guidance module based on dynamic convolution, in which an efficient convolution factorization is proposed to simultaneously reduce its complexity and progressively model high-frequency structures. Extensive experiments show that our method achieves superior or competitive results on KITTI benchmark and NYUv2 dataset.

Zhenyu Zhang, Xuxi Chen, Tianlong Chen et al.

The lottery ticket hypothesis (LTH) reveals the existence of winning tickets (sparse but critical subnetworks) for dense networks, that can be trained in isolation from random initialization to match the latter's accuracies. However, finding winning tickets requires burdensome computations in the train-prune-retrain process, especially on large-scale datasets (e.g., ImageNet), restricting their practical benefits. This paper explores a new perspective on finding lottery tickets more efficiently, by doing so only with a specially selected subset of data, called Pruning-Aware Critical set (PrAC set), rather than using the full training set. The concept of PrAC set was inspired by the recent observation, that deep networks have samples that are either hard to memorize during training, or easy to forget during pruning. A PrAC set is thus hypothesized to capture those most challenging and informative examples for the dense model. We observe that a high-quality winning ticket can be found with training and pruning the dense network on the very compact PrAC set, which can substantially save training iterations for the ticket finding process. Extensive experiments validate our proposal across diverse datasets and network architectures. Specifically, on CIFAR-10, CIFAR-100, and Tiny ImageNet, we locate effective PrAC sets at 35.32%~78.19% of their training set sizes. On top of them, we can obtain the same competitive winning tickets for the corresponding dense networks, yet saving up to 82.85%~92.77%, 63.54%~74.92%, and 76.14%~86.56% training iterations, respectively. Crucially, we show that a PrAC set found is reusable across different network architectures, which can amortize the extra cost of finding PrAC sets, yielding a practical regime for efficient lottery ticket finding.

Zhenyu Zhang, Stéphane Lathuilière, Andrea Pilzer et al.

In this work, we tackle the problem of online adaptation for stereo depth estimation, that consists in continuously adapting a deep network to a target video recordedin an environment different from that of the source training set. To address this problem, we propose a novel Online Meta-Learning model with Adaption (OMLA). Our proposal is based on two main contributions. First, to reducethe domain-shift between source and target feature distributions we introduce an online feature alignment procedurederived from Batch Normalization. Second, we devise a meta-learning approach that exploits feature alignment forfaster convergence in an online learning setting. Additionally, we propose a meta-pre-training algorithm in order toobtain initial network weights on the source dataset whichfacilitate adaptation on future data streams. Experimentally, we show that both OMLA and meta-pre-training helpthe model to adapt faster to a new environment. Our proposal is evaluated on the wellestablished KITTI dataset,where we show that our online method is competitive withstate of the art algorithms trained in a batch setting.