Zhengqiang Zhang

Rongyuan Wu, Tao Yang, Lingchen Sun et al.

Owe to the powerful generative priors, the pre-trained text-to-image (T2I) diffusion models have become increasingly popular in solving the real-world image super-resolution problem. However, as a consequence of the heavy quality degradation of input low-resolution (LR) images, the destruction of local structures can lead to ambiguous image semantics. As a result, the content of reproduced high-resolution image may have semantic errors, deteriorating the super-resolution performance. To address this issue, we present a semantics-aware approach to better preserve the semantic fidelity of generative real-world image super-resolution. First, we train a degradation-aware prompt extractor, which can generate accurate soft and hard semantic prompts even under strong degradation. The hard semantic prompts refer to the image tags, aiming to enhance the local perception ability of the T2I model, while the soft semantic prompts compensate for the hard ones to provide additional representation information. These semantic prompts encourage the T2I model to generate detailed and semantically accurate results. Furthermore, during the inference process, we integrate the LR images into the initial sampling noise to mitigate the diffusion model's tendency to generate excessive random details. The experiments show that our method can reproduce more realistic image details and hold better the semantics. The source code of our method can be found at https://github.com/cswry/SeeSR.

Du Chen, Zhengqiang Zhang, Jie Liang et al.

Generative adversarial networks (GAN) and generative diffusion models (DM) have been widely used in real-world image super-resolution (Real-ISR) to enhance the image perceptual quality. However, these generative models are prone to generating visual artifacts and false image structures, resulting in unnatural Real-ISR results. Based on the fact that natural images exhibit high self-similarities, i.e., a local patch can have many similar patches to it in the whole image, in this work we propose a simple yet effective self-similarity loss (SSL) to improve the performance of generative Real-ISR models, enhancing the hallucination of structural and textural details while reducing the unpleasant visual artifacts. Specifically, we compute a self-similarity graph (SSG) of the ground-truth image, and enforce the SSG of Real-ISR output to be close to it. To reduce the training cost and focus on edge areas, we generate an edge mask from the ground-truth image, and compute the SSG only on the masked pixels. The proposed SSL serves as a general plug-and-play penalty, which could be easily applied to the off-the-shelf Real-ISR models. Our experiments demonstrate that, by coupling with SSL, the performance of many state-of-the-art Real-ISR models, including those GAN and DM based ones, can be largely improved, reproducing more perceptually realistic image details and eliminating many false reconstructions and visual artifacts. Codes and supplementary material can be found at https://github.com/ChrisDud0257/SSL

Chaodong Xiao, Zhengqiang Zhang, Lei Zhang

Transformers have achieved widespread and remarkable success, while the computational complexity of their attention modules remains a major bottleneck for vision tasks. Existing methods mainly employ 8-bit or 4-bit quantization to balance efficiency and accuracy. In this paper, with theoretical justification, we indicate that binarization of attention preserves the essential similarity relationships, and propose BinaryAttention, an effective method for fast and accurate 1-bit qk-attention. Specifically, we retain only the sign of queries and keys in computing the attention, and replace the floating dot products with bit-wise operations, significantly reducing the computational cost. We mitigate the inherent information loss under 1-bit quantization by incorporating a learnable bias, and enable end-to-end acceleration. To maintain the accuracy of attention, we adopt quantization-aware training and self-distillation techniques, mitigating quantization errors while ensuring sign-aligned similarity. BinaryAttention is more than 2x faster than FlashAttention2 on A100 GPUs. Extensive experiments on vision transformer and diffusion transformer benchmarks demonstrate that BinaryAttention matches or even exceeds full-precision attention, validating its effectiveness. Our work provides a highly efficient and effective alternative to full-precision attention, pushing the frontier of low-bit vision and diffusion transformers. The codes and models can be found at https://github.com/EdwardChasel/BinaryAttention.

Ruihuang Li, Zhengqiang Zhang, Chenhang He et al.

Recent vision-language pre-training models have exhibited remarkable generalization ability in zero-shot recognition tasks. Previous open-vocabulary 3D scene understanding methods mostly focus on training 3D models using either image or text supervision while neglecting the collective strength of all modalities. In this work, we propose a Dense Multimodal Alignment (DMA) framework to densely co-embed different modalities into a common space for maximizing their synergistic benefits. Instead of extracting coarse view- or region-level text prompts, we leverage large vision-language models to extract complete category information and scalable scene descriptions to build the text modality, and take image modality as the bridge to build dense point-pixel-text associations. Besides, in order to enhance the generalization ability of the 2D model for downstream 3D tasks without compromising the open-vocabulary capability, we employ a dual-path integration approach to combine frozen CLIP visual features and learnable mask features. Extensive experiments show that our DMA method produces highly competitive open-vocabulary segmentation performance on various indoor and outdoor tasks.

Lingchen Sun, Rongyuan Wu, Jie Liang et al.

The generative priors of pre-trained latent diffusion models (DMs) have demonstrated great potential to enhance the visual quality of image super-resolution (SR) results. However, the noise sampling process in DMs introduces randomness in the SR outputs, and the generated contents can differ a lot with different noise samples. The multi-step diffusion process can be accelerated by distilling methods, but the generative capacity is difficult to control. To address these issues, we analyze the respective advantages of DMs and generative adversarial networks (GANs) and propose to partition the generative SR process into two stages, where the DM is employed for reconstructing image structures and the GAN is employed for improving fine-grained details. Specifically, we propose a non-uniform timestep sampling strategy in the first stage. A single timestep sampling is first applied to extract the coarse information from the input image, then a few reverse steps are used to reconstruct the main structures. In the second stage, we finetune the decoder of the pre-trained variational auto-encoder by adversarial GAN training for deterministic detail enhancement. Once trained, our proposed method, namely content consistent super-resolution (CCSR),allows flexible use of different diffusion steps in the inference stage without re-training. Extensive experiments show that with 2 or even 1 diffusion step, CCSR can significantly improve the content consistency of SR outputs while keeping high perceptual quality. Codes and models can be found at \href{https://github.com/csslc/CCSR}{https://github.com/csslc/CCSR}.

Zhengqiang Zhang, Ruihuang Li, Lei Zhang

While image generation with diffusion models has achieved a great success, generating images of higher resolution than the training size remains a challenging task due to the high computational cost. Current methods typically perform the entire sampling process at full resolution and process all frequency components simultaneously, contradicting with the inherent coarse-to-fine nature of latent diffusion models and wasting computations on processing premature high-frequency details at early diffusion stages. To address this issue, we introduce an efficient $\textbf{Fre}$quency-aware $\textbf{Ca}$scaded $\textbf{S}$ampling framework, $\textbf{FreCaS}$ in short, for higher-resolution image generation. FreCaS decomposes the sampling process into cascaded stages with gradually increased resolutions, progressively expanding frequency bands and refining the corresponding details. We propose an innovative frequency-aware classifier-free guidance (FA-CFG) strategy to assign different guidance strengths for different frequency components, directing the diffusion model to add new details in the expanded frequency domain of each stage. Additionally, we fuse the cross-attention maps of previous and current stages to avoid synthesizing unfaithful layouts. Experiments demonstrate that FreCaS significantly outperforms state-of-the-art methods in image quality and generation speed. In particular, FreCaS is about 2.86$\times$ and 6.07$\times$ faster than ScaleCrafter and DemoFusion in generating a 2048$\times$2048 image using a pre-trained SDXL model and achieves an FID$_b$ improvement of 11.6 and 3.7, respectively. FreCaS can be easily extended to more complex models such as SD3. The source code of FreCaS can be found at https://github.com/xtudbxk/FreCaS.

Zhengqiang Zhang, Ruihuang Li, Shi Guo et al.

Online video super-resolution (online-VSR) highly relies on an effective alignment module to aggregate temporal information, while the strict latency requirement makes accurate and efficient alignment very challenging. Though much progress has been achieved, most of the existing online-VSR methods estimate the motion fields of each frame separately to perform alignment, which is computationally redundant and ignores the fact that the motion fields of adjacent frames are correlated. In this work, we propose an efficient Temporal Motion Propagation (TMP) method, which leverages the continuity of motion field to achieve fast pixel-level alignment among consecutive frames. Specifically, we first propagate the offsets from previous frames to the current frame, and then refine them in the neighborhood, which significantly reduces the matching space and speeds up the offset estimation process. Furthermore, to enhance the robustness of alignment, we perform spatial-wise weighting on the warped features, where the positions with more precise offsets are assigned higher importance. Experiments on benchmark datasets demonstrate that the proposed TMP method achieves leading online-VSR accuracy as well as inference speed. The source code of TMP can be found at https://github.com/xtudbxk/TMP.

Zhengqiang Zhang, Rongyuan Wu, Lingchen Sun et al.

Effective and efficient tokenization plays an important role in image representation and generation. Conventional methods, constrained by uniform 2D/1D grid tokenization, are inflexible to represent regions with varying shapes and textures and at different locations, limiting their efficacy of feature representation. In this work, we propose $\textbf{GPSToken}$, a novel $\textbf{G}$aussian $\textbf{P}$arameterized $\textbf{S}$patially-adaptive $\textbf{Token}$ization framework, to achieve non-uniform image tokenization by leveraging parametric 2D Gaussians to dynamically model the shape, position, and textures of different image regions. We first employ an entropy-driven algorithm to partition the image into texture-homogeneous regions of variable sizes. Then, we parameterize each region as a 2D Gaussian (mean for position, covariance for shape) coupled with texture features. A specialized transformer is trained to optimize the Gaussian parameters, enabling continuous adaptation of position/shape and content-aware feature extraction. During decoding, Gaussian parameterized tokens are reconstructed into 2D feature maps through a differentiable splatting-based renderer, bridging our adaptive tokenization with standard decoders for end-to-end training. GPSToken disentangles spatial layout (Gaussian parameters) from texture features to enable efficient two-stage generation: structural layout synthesis using lightweight networks, followed by structure-conditioned texture generation. Experiments demonstrate the state-of-the-art performance of GPSToken, which achieves rFID and FID scores of 0.65 and 1.50 on image reconstruction and generation tasks using 128 tokens, respectively. Codes and models of GPSToken can be found at $\href{https://github.com/xtudbxk/GPSToken}{https://github.com/xtudbxk/GPSToken}$.

Shi Guo, Jianqi Ma, Xi Yang et al.

Denoising and demosaicking are two fundamental steps in reconstructing a clean full-color video from raw data, while performing video denoising and demosaicking jointly, namely VJDD, could lead to better video restoration performance than performing them separately. In addition to restoration accuracy, another key challenge to VJDD lies in the temporal consistency of consecutive frames. This issue exacerbates when perceptual regularization terms are introduced to enhance video perceptual quality. To address these challenges, we present a new VJDD framework by consistent and accurate latent space propagation, which leverages the estimation of previous frames as prior knowledge to ensure consistent recovery of the current frame. A data temporal consistency (DTC) loss and a relational perception consistency (RPC) loss are accordingly designed. Compared with the commonly used flow-based losses, the proposed losses can circumvent the error accumulation problem caused by inaccurate flow estimation and effectively handle intensity changes in videos, improving much the temporal consistency of output videos while preserving texture details. Extensive experiments demonstrate the leading VJDD performance of our method in term of restoration accuracy, perceptual quality and temporal consistency. Codes and dataset are available at \url{https://github.com/GuoShi28/VJDD}.

Jie Liang, Radu Timofte, Qiaosi Yi et al.

In this paper, we present a comprehensive overview of the NTIRE 2025 challenge on the 2nd Restore Any Image Model (RAIM) in the Wild. This challenge established a new benchmark for real-world image restoration, featuring diverse scenarios with and without reference ground truth. Participants were tasked with restoring real-captured images suffering from complex and unknown degradations, where both perceptual quality and fidelity were critically evaluated. The challenge comprised two tracks: (1) the low-light joint denoising and demosaicing (JDD) task, and (2) the image detail enhancement/generation task. Each track included two sub-tasks. The first sub-task involved paired data with available ground truth, enabling quantitative evaluation. The second sub-task dealt with real-world yet unpaired images, emphasizing restoration efficiency and subjective quality assessed through a comprehensive user study. In total, the challenge attracted nearly 300 registrations, with 51 teams submitting more than 600 results. The top-performing methods advanced the state of the art in image restoration and received unanimous recognition from all 20+ expert judges. The datasets used in Track 1 and Track 2 are available at https://drive.google.com/drive/folders/1Mgqve-yNcE26IIieI8lMIf-25VvZRs_J and https://drive.google.com/drive/folders/1UB7nnzLwqDZOwDmD9aT8J0KVg2ag4Qae, respectively. The official challenge pages for Track 1 and Track 2 can be found at https://codalab.lisn.upsaclay.fr/competitions/21334#learn_the_details and https://codalab.lisn.upsaclay.fr/competitions/21623#learn_the_details.

Ruihuang Li, Liyi Chen, Zhengqiang Zhang et al.

Text-based 2D diffusion models have demonstrated impressive capabilities in image generation and editing. Meanwhile, the 2D diffusion models also exhibit substantial potentials for 3D editing tasks. However, how to achieve consistent edits across multiple viewpoints remains a challenge. While the iterative dataset update method is capable of achieving global consistency, it suffers from slow convergence and over-smoothed textures. We propose SyncNoise, a novel geometry-guided multi-view consistent noise editing approach for high-fidelity 3D scene editing. SyncNoise synchronously edits multiple views with 2D diffusion models while enforcing multi-view noise predictions to be geometrically consistent, which ensures global consistency in both semantic structure and low-frequency appearance. To further enhance local consistency in high-frequency details, we set a group of anchor views and propagate them to their neighboring frames through cross-view reprojection. To improve the reliability of multi-view correspondences, we introduce depth supervision during training to enhance the reconstruction of precise geometries. Our method achieves high-quality 3D editing results respecting the textual instructions, especially in scenes with complex textures, by enhancing geometric consistency at the noise and pixel levels.

Yang Cao, Zhengqiang Zhang, Enze Xie et al.

Detecting transparent objects in natural scenes is challenging due to the low contrast in texture, brightness and colors. Recent deep-learning-based works reveal that it is effective to leverage boundaries for transparent object detection (TOD). However, these methods usually encounter boundary-related imbalance problem, leading to limited generation capability. Detailly, a kind of boundaries in the background, which share the same characteristics with boundaries of transparent objects but have much smaller amounts, usually hurt the performance. To conquer the boundary-related imbalance problem, we propose a novel content-dependent data augmentation method termed FakeMix. Considering collecting these trouble-maker boundaries in the background is hard without corresponding annotations, we elaborately generate them by appending the boundaries of transparent objects from other samples into the current image during training, which adjusts the data space and improves the generalization of the models. Further, we present AdaptiveASPP, an enhanced version of ASPP, that can capture multi-scale and cross-modality features dynamically. Extensive experiments demonstrate that our methods clearly outperform the state-of-the-art methods. We also show that our approach can also transfer well on related tasks, in which the model meets similar troubles, such as mirror detection, glass detection, and camouflaged object detection. Code will be made publicly available.

Zhengqiang Zhang, Shujian Yu, Shi Yin et al.

Weakly-supervised semantic segmentation aims to assign each pixel a semantic category under weak supervisions, such as image-level tags. Most of existing weakly-supervised semantic segmentation methods do not use any feedback from segmentation output and can be considered as open-loop systems. They are prone to accumulated errors because of the static seeds and the sensitive structure information. In this paper, we propose a generic self-adaptation mechanism for existing weakly-supervised semantic segmentation methods by introducing two feedback chains, thus constituting a closed-loop system. Specifically, the first chain iteratively produces dynamic seeds by incorporating cross-image structure information, whereas the second chain further expands seed regions by a customized random walk process to reconcile inner-image structure information characterized by superpixels. Experiments on PASCAL VOC 2012 suggest that our network outperforms state-of-the-art methods with significantly less computational and memory burden.

Shi Yin, Zhengqiang Zhang, Qinmu Peng et al.

High angular resolution diffusion imaging (HARDI) demands a lager amount of data measurements compared to diffusion tensor imaging, restricting its use in practice. In this work, we explore a learning-based approach to reconstruct HARDI from a smaller number of measurements in q-space. The approach aims to directly learn the mapping relationship between the measured and HARDI signals from the collecting HARDI acquisitions of other subjects. Specifically, the mapping is represented as a 1D encoder-decoder convolutional neural network under the guidance of the compressed sensing (CS) theory for HARDI reconstruction. The proposed network architecture mainly consists of two parts: an encoder network produces the sparse coefficients and a decoder network yields a reconstruction result. Experiment results demonstrate we can robustly reconstruct HARDI signals with the accurate results and fast speed.

Shi Yin, Zhengqiang Zhang, Hongming Li et al.

It remains challenging to automatically segment kidneys in clinical ultrasound images due to the kidneys' varied shapes and image intensity distributions, although semi-automatic methods have achieved promising performance. In this study, we developed a novel boundary distance regression deep neural network to segment the kidneys, informed by the fact that the kidney boundaries are relatively consistent across images in terms of their appearance. Particularly, we first use deep neural networks pre-trained for classification of natural images to extract high-level image features from ultrasound images, then these feature maps are used as input to learn kidney boundary distance maps using a boundary distance regression network, and finally the predicted boundary distance maps are classified as kidney pixels or non-kidney pixels using a pixel classification network in an end-to-end learning fashion. Experimental results have demonstrated that our method could effectively improve the performance of automatic kidney segmentation, significantly better than deep learning based pixel classification networks.

Shi Yin, Qinmu Peng, Hongming Li et al.

It remains challenging to automatically segment kidneys in clinical ultrasound (US) images due to the kidneys' varied shapes and image intensity distributions, although semi-automatic methods have achieved promising performance. In this study, we propose subsequent boundary distance regression and pixel classification networks to segment the kidneys, informed by the fact that the kidney boundaries have relatively homogenous texture patterns across images. Particularly, we first use deep neural networks pre-trained for classification of natural images to extract high-level image features from US images, then these features are used as input to learn kidney boundary distance maps using a boundary distance regression network, and finally the predicted boundary distance maps are classified as kidney pixels or non-kidney pixels using a pixel classification network in an end-to-end learning fashion. We also adopted a data-augmentation method based on kidney shape registration to generate enriched training data from a small number of US images with manually segmented kidney labels. Experimental results have demonstrated that our method could effectively improve the performance of automatic kidney segmentation, significantly better than deep learning-based pixel classification networks.