arXiv subjects

Zhengjun Zha

Publications and source records attributed to Zhengjun Zha.

At least 19 recordsLinked to original sources

arXiv2026

Follow Your Track: Precise Skeleton Animation Controlled by 3D Trajectories

4D generation aims to animate 3D objects with realistic motion, holding great promise for applications. Existing methods typically decouple 3D asset generation from motion synthesis: acquire a 3D asset, prepare a structural representation like mesh and Gaussians, and synthesize motion from text or video control signals. However, dense mesh and Gaussian representations incur high computational costs and are prone to temporal artifacts, limiting animation quality and duration to only short clips. Meanwhile, text lacks fine-grained spatial and temporal details such as timing and coordination, while video entangles motion with appearance and background. Together, these limitations result in 4D animations that suffer from poor temporal consistency, wrong identification, and limited controllability. We address these issues with \texttt{ACT}, a trajectory-conditioned framework for topology-general skeletal animation. ACT uses skeletons as a compact structured and compute-efficient representation and 3D point trajectories from monocular video as explicit motion guidance which provide detailed motion patterns without appearance entanglement. At the core of ACT is a Routed Trajectory Injector, which achieves accurate and robust trajectory-to-joint transfer through three complementary designs: prior-guided hard routing establishes precise skeleton-to-mesh correspondences, global routing enables holistic joint-track interaction for full-body motion awareness, and local windowed cross-attention enforces fine-grained temporal alignment, improving micro-timing and reducing motion misalignment across varying motion rates. Extensive experiments demonstrate that \texttt{ACT} significantly outperforms existing methods in fidelity and temporal consistency.

cs.CV↗

arXiv2026

Tail-Aware HiFloat4: W4A4 Post-Training Quantization for Wan2.2

This report describes Tail-Aware HiFloat4, our submission to the low-bit text-to-video generation quantization challenge. Our method adapts the public ViDiT-Q post-training quantization pipeline to Wan2.2 under the HiFloat4 numerical format. We quantize the main linear layers in both Wan2.2 transformer modules with W4A4 HiFloat4 fake quantization, keep numerically sensitive boundary modules in high precision, and introduce an activation-tail-aware percentile calibration module for channel-mask construction. Together with compact PTQ-state restoration, this design reduces the influence of rare calibration outliers while keeping the runtime HiFloat4 arithmetic and sampling pipeline unchanged.

cs.AI↗

arXiv2026

GS-STVSR: Ultra-Efficient Continuous Spatio-Temporal Video Super-Resolution via 2D Gaussian Splatting

Continuous Spatio-Temporal Video Super-Resolution (C-STVSR) aims to simultaneously enhance the spatial resolution and frame rate of videos by arbitrary scale factors, offering greater flexibility than fixed-scale methods that are constrained by predefined upsampling ratios. In recent years, methods based on Implicit Neural Representations (INR) have made significant progress in C-STVSR by learning continuous mappings from spatio-temporal coordinates to pixel values. However, these methods fundamentally rely on dense pixel-wise grid queries, causing computational cost to scale linearly with the number of interpolated frames and severely limiting inference efficiency. We propose GS-STVSR, an ultra-efficient C-STVSR framework based on 2D Gaussian Splatting (2D-GS) that drives the spatiotemporal evolution of Gaussian kernels through continuous motion modeling, bypassing dense grid queries entirely. We exploit the strong temporal stability of covariance parameters for lightweight intermediate fitting, design an optical flow-guided motion module to derive Gaussian position and color at arbitrary time steps, introduce a Covariance resampling alignment module to prevent covariance drift, and propose an adaptive offset window for large-scale motion. Extensive experiments on Vid4, GoPro, and Adobe240 show that GS-STVSR achieves state-of-the-art quality across all benchmarks. Moreover, its inference time remains nearly constant at conventional temporal scales (X2--X8) and delivers over X3 speedup at extreme scales X32, demonstrating strong practical applicability.

cs.CV↗

arXiv2026

The First Challenge on Mobile Real-World Image Super-Resolution at NTIRE 2026: Benchmark Results and Method Overview

This paper provides a review of the NTIRE 2026 challenge on mobile real-world image super-resolution, highlighting the proposed solutions and the resulting outcomes. The challenge aims to recover high-resolution (HR) images from low-resolution (LR) counterparts generated through unknown degradations with a x4 scaling factor while ensuring the models remain executable on mobile devices. The objective is to develop effective and efficient network designs or solutions that achieve state-of-the-art real-world image super-resolution performance. The track of the challenge evaluates performance using a weighted combination of image quality assessment (IQA) score and speedup ratios. The competition attracted 108 registrants, with 16 teams achieving a valid score in the final ranking. This collaborative effort advances the performance of mobile real-world image super-resolution while offering an in-depth overview of the latest trends in the field.

cs.CV↗

arXiv2025

Iterative Inference-time Scaling with Adaptive Frequency Steering for Image Super-Resolution

Diffusion models have become a leading paradigm for image super-resolution (SR), but existing methods struggle to guarantee both the high-frequency perceptual quality and the low-frequency structural fidelity of generated images. Although inference-time scaling can theoretically improve this trade-off by allocating more computation, existing strategies remain suboptimal: reward-driven particle optimization often causes perceptual over-smoothing, while optimal-path search tends to lose structural consistency. To overcome these difficulties, we propose Iterative Diffusion Inference-Time Scaling with Adaptive Frequency Steering (IAFS), a training-free framework that jointly leverages iterative refinement and frequency-aware particle fusion. IAFS addresses the challenge of balancing perceptual quality and structural fidelity by progressively refining the generated image through iterative correction of structural deviations. Simultaneously, it ensures effective frequency fusion by adaptively integrating high-frequency perceptual cues with low-frequency structural information, allowing for a more accurate and balanced reconstruction across different image details. Extensive experiments across multiple diffusion-based SR models show that IAFS effectively resolves the perception-fidelity conflict, yielding consistently improved perceptual detail and structural accuracy, and outperforming existing inference-time scaling methods.

cs.CV↗

arXiv2025

Beyond Sunk Costs: Boosting LLM Pre-training Efficiency via Orthogonal Growth of Mixture-of-Experts

As the computational demands for pre-training Large Language Models (LLMs) continue to surge, the need for efficient training paradigms becomes critical. Despite the vast resources already invested in existing pre-trained checkpoints, these assets often remain under-leveraged due to architectural limitations. We introduce an "orthogonal growth" strategy designed to "recycle" these checkpoints by strategically expanding their parameters prior to continued training. Our method focuses on optimizing converged Mixture-of-Experts (MoE) models through two dimensions: interpositional layer copying for increased depth and noisy expert duplication for expanded width. Through extensive scaling laws analysis, we demonstrate a strong positive correlation between the "sunk cost" (prior investment) and the final model accuracy. Empirical results on models up to 70B parameters and 1T tokens show that our recycling approach yields a 10.6% accuracy improvement compared to training from scratch under identical extra compute budgets. This work provides a cost-effective blueprint for sustainable large-scale LLM development.

cs.LG↗

arXiv2025

GRACE: Estimating Geometry-level 3D Human-Scene Contact from 2D Images

Estimating the geometry level of human-scene contact aims to ground specific contact surface points at 3D human geometries, which provides a spatial prior and bridges the interaction between human and scene, supporting applications such as human behavior analysis, embodied AI, and AR/VR. To complete the task, existing approaches predominantly rely on parametric human models (e.g., SMPL), which establish correspondences between images and contact regions through fixed SMPL vertex sequences. This actually completes the mapping from image features to an ordered sequence. However, this approach lacks consideration of geometry, limiting its generalizability in distinct human geometries. In this paper, we introduce GRACE (Geometry-level Reasoning for 3D Human-scene Contact Estimation), a new paradigm for 3D human contact estimation. GRACE incorporates a point cloud encoder-decoder architecture along with a hierarchical feature extraction and fusion module, enabling the effective integration of 3D human geometric structures with 2D interaction semantics derived from images. Guided by visual cues, GRACE establishes an implicit mapping from geometric features to the vertex space of the 3D human mesh, thereby achieving accurate modeling of contact regions. This design ensures high prediction accuracy and endows the framework with strong generalization capability across diverse human geometries. Extensive experiments on multiple benchmark datasets demonstrate that GRACE achieves state-of-the-art performance in contact estimation, with additional results further validating its robust generalization to unstructured human point clouds.

cs.CV↗

arXiv2025

EMoTive: Event-guided Trajectory Modeling for 3D Motion Estimation

Visual 3D motion estimation aims to infer the motion of 2D pixels in 3D space based on visual cues. The key challenge arises from depth variation induced spatio-temporal motion inconsistencies, disrupting the assumptions of local spatial or temporal motion smoothness in previous motion estimation frameworks. In contrast, event cameras offer new possibilities for 3D motion estimation through continuous adaptive pixel-level responses to scene changes. This paper presents EMoTive, a novel event-based framework that models spatio-temporal trajectories via event-guided non-uniform parametric curves, effectively characterizing locally heterogeneous spatio-temporal motion. Specifically, we first introduce Event Kymograph - an event projection method that leverages a continuous temporal projection kernel and decouples spatial observations to encode fine-grained temporal evolution explicitly. For motion representation, we introduce a density-aware adaptation mechanism to fuse spatial and temporal features under event guidance, coupled with a non-uniform rational curve parameterization framework to adaptively model heterogeneous trajectories. The final 3D motion estimation is achieved through multi-temporal sampling of parametric trajectories, yielding optical flow and depth motion fields. To facilitate evaluation, we introduce CarlaEvent3D, a multi-dynamic synthetic dataset for comprehensive validation. Extensive experiments on both this dataset and a real-world benchmark demonstrate the effectiveness of the proposed method.

cs.CV↗

arXiv2025

PFSD: A Multi-Modal Pedestrian-Focus Scene Dataset for Rich Tasks in Semi-Structured Environments

Recent advancements in autonomous driving perception have revealed exceptional capabilities within structured environments dominated by vehicular traffic. However, current perception models exhibit significant limitations in semi-structured environments, where dynamic pedestrians with more diverse irregular movement and occlusion prevail. We attribute this shortcoming to the scarcity of high-quality datasets in semi-structured scenes, particularly concerning pedestrian perception and prediction. In this work, we present the multi-modal Pedestrian-Focused Scene Dataset(PFSD), rigorously annotated in semi-structured scenes with the format of nuScenes. PFSD provides comprehensive multi-modal data annotations with point cloud segmentation, detection, and object IDs for tracking. It encompasses over 130,000 pedestrian instances captured across various scenarios with varying densities, movement patterns, and occlusions. Furthermore, to demonstrate the importance of addressing the challenges posed by more diverse and complex semi-structured environments, we propose a novel Hybrid Multi-Scale Fusion Network (HMFN). Specifically, to detect pedestrians in densely populated and occluded scenarios, our method effectively captures and fuses multi-scale features using a meticulously designed hybrid framework that integrates sparse and vanilla convolutions. Extensive experiments on PFSD demonstrate that HMFN attains improvement in mean Average Precision (mAP) over existing methods, thereby underscoring its efficacy in addressing the challenges of 3D pedestrian detection in complex semi-structured environments. Coding and benchmark are available.

cs.CV↗

arXiv2025

Optimizing Large Language Model Training Using FP4 Quantization

The growing computational demands of training large language models (LLMs) necessitate more efficient methods. Quantized training presents a promising solution by enabling low-bit arithmetic operations to reduce these costs. While FP8 precision has demonstrated feasibility, leveraging FP4 remains a challenge due to significant quantization errors and limited representational capacity. This work introduces the first FP4 training framework for LLMs, addressing these challenges with two key innovations: a differentiable quantization estimator for precise weight updates and an outlier clamping and compensation strategy to prevent activation collapse. To ensure stability, the framework integrates a mixed-precision training scheme and vector-wise quantization. Experimental results demonstrate that our FP4 framework achieves accuracy comparable to BF16 and FP8, with minimal degradation, scaling effectively to 13B-parameter LLMs trained on up to 100B tokens. With the emergence of next-generation hardware supporting FP4, our framework sets a foundation for efficient ultra-low precision training.

cs.LG↗

arXiv2024

I2V-Adapter: A General Image-to-Video Adapter for Diffusion Models

Text-guided image-to-video (I2V) generation aims to generate a coherent video that preserves the identity of the input image and semantically aligns with the input prompt. Existing methods typically augment pretrained text-to-video (T2V) models by either concatenating the image with noised video frames channel-wise before being fed into the model or injecting the image embedding produced by pretrained image encoders in cross-attention modules. However, the former approach often necessitates altering the fundamental weights of pretrained T2V models, thus restricting the model's compatibility within the open-source communities and disrupting the model's prior knowledge. Meanwhile, the latter typically fails to preserve the identity of the input image. We present I2V-Adapter to overcome such limitations. I2V-Adapter adeptly propagates the unnoised input image to subsequent noised frames through a cross-frame attention mechanism, maintaining the identity of the input image without any changes to the pretrained T2V model. Notably, I2V-Adapter only introduces a few trainable parameters, significantly alleviating the training cost and also ensures compatibility with existing community-driven personalized models and control tools. Moreover, we propose a novel Frame Similarity Prior to balance the motion amplitude and the stability of generated videos through two adjustable control coefficients. Our experimental results demonstrate that I2V-Adapter is capable of producing high-quality videos. This performance, coupled with its agility and adaptability, represents a substantial advancement in the field of I2V, particularly for personalized and controllable applications.

cs.CV↗

arXiv2024

The Ninth NTIRE 2024 Efficient Super-Resolution Challenge Report

This paper provides a comprehensive review of the NTIRE 2024 challenge, focusing on efficient single-image super-resolution (ESR) solutions and their outcomes. The task of this challenge is to super-resolve an input image with a magnification factor of x4 based on pairs of low and corresponding high-resolution images. The primary objective is to develop networks that optimize various aspects such as runtime, parameters, and FLOPs, while still maintaining a peak signal-to-noise ratio (PSNR) of approximately 26.90 dB on the DIV2K_LSDIR_valid dataset and 26.99 dB on the DIV2K_LSDIR_test dataset. In addition, this challenge has 4 tracks including the main track (overall performance), sub-track 1 (runtime), sub-track 2 (FLOPs), and sub-track 3 (parameters). In the main track, all three metrics (ie runtime, FLOPs, and parameter count) were considered. The ranking of the main track is calculated based on a weighted sum-up of the scores of all other sub-tracks. In sub-track 1, the practical runtime performance of the submissions was evaluated, and the corresponding score was used to determine the ranking. In sub-track 2, the number of FLOPs was considered. The score calculated based on the corresponding FLOPs was used to determine the ranking. In sub-track 3, the number of parameters was considered. The score calculated based on the corresponding parameters was used to determine the ranking. RLFN is set as the baseline for efficiency measurement. The challenge had 262 registered participants, and 34 teams made valid submissions. They gauge the state-of-the-art in efficient single-image super-resolution. To facilitate the reproducibility of the challenge and enable other researchers to build upon these findings, the code and the pre-trained model of validated solutions are made publicly available at https://github.com/Amazingren/NTIRE2024_ESR/.

cs.CV↗

arXiv2024

Real-Time 4K Super-Resolution of Compressed AVIF Images. AIS 2024 Challenge Survey

This paper introduces a novel benchmark as part of the AIS 2024 Real-Time Image Super-Resolution (RTSR) Challenge, which aims to upscale compressed images from 540p to 4K resolution (4x factor) in real-time on commercial GPUs. For this, we use a diverse test set containing a variety of 4K images ranging from digital art to gaming and photography. The images are compressed using the modern AVIF codec, instead of JPEG. All the proposed methods improve PSNR fidelity over Lanczos interpolation, and process images under 10ms. Out of the 160 participants, 25 teams submitted their code and models. The solutions present novel designs tailored for memory-efficiency and runtime on edge devices. This survey describes the best solutions for real-time SR of compressed high-resolution images.

cs.CV↗

arXiv2023

Learning to Dub Movies via Hierarchical Prosody Models

Given a piece of text, a video clip and a reference audio, the movie dubbing (also known as visual voice clone V2C) task aims to generate speeches that match the speaker's emotion presented in the video using the desired speaker voice as reference. V2C is more challenging than conventional text-to-speech tasks as it additionally requires the generated speech to exactly match the varying emotions and speaking speed presented in the video. Unlike previous works, we propose a novel movie dubbing architecture to tackle these problems via hierarchical prosody modelling, which bridges the visual information to corresponding speech prosody from three aspects: lip, face, and scene. Specifically, we align lip movement to the speech duration, and convey facial expression to speech energy and pitch via attention mechanism based on valence and arousal representations inspired by recent psychology findings. Moreover, we design an emotion booster to capture the atmosphere from global video scenes. All these embeddings together are used to generate mel-spectrogram and then convert to speech waves via existing vocoder. Extensive experimental results on the Chem and V2C benchmark datasets demonstrate the favorable performance of the proposed method. The source code and trained models will be released to the public.

cs.CL↗

arXiv2022

Few-shot Learning with Global Relatedness Decoupled-Distillation

Despite the success that metric learning based approaches have achieved in few-shot learning, recent works reveal the ineffectiveness of their episodic training mode. In this paper, we point out two potential reasons for this problem: 1) the random episodic labels can only provide limited supervision information, while the relatedness information between the query and support samples is not fully exploited; 2) the meta-learner is usually constrained by the limited contextual information of the local episode. To overcome these problems, we propose a new Global Relatedness Decoupled-Distillation (GRDD) method using the global category knowledge and the Relatedness Decoupled-Distillation (RDD) strategy. Our GRDD learns new visual concepts quickly by imitating the habit of humans, i.e. learning from the deep knowledge distilled from the teacher. More specifically, we first train a global learner on the entire base subset using category labels as supervision to leverage the global context information of the categories. Then, the well-trained global learner is used to simulate the query-support relatedness in global dependencies. Finally, the distilled global query-support relatedness is explicitly used to train the meta-learner using the RDD strategy, with the goal of making the meta-learner more discriminative. The RDD strategy aims to decouple the dense query-support relatedness into the groups of sparse decoupled relatedness. Moreover, only the relatedness of a single support sample with other query samples is considered in each group. By distilling the sparse decoupled relatedness group by group, sharper relatedness can be effectively distilled to the meta-learner, thereby facilitating the learning of a discriminative meta-learner. We conduct extensive experiments on the miniImagenet and CIFAR-FS datasets, which show the state-of-the-art performance of our GRDD method.

cs.CV↗

arXiv2021

Weakly Supervised High-Fidelity Clothing Model Generation

The development of online economics arouses the demand of generating images of models on product clothes, to display new clothes and promote sales. However, the expensive proprietary model images challenge the existing image virtual try-on methods in this scenario, as most of them need to be trained on considerable amounts of model images accompanied with paired clothes images. In this paper, we propose a cheap yet scalable weakly-supervised method called Deep Generative Projection (DGP) to address this specific scenario. Lying in the heart of the proposed method is to imitate the process of human predicting the wearing effect, which is an unsupervised imagination based on life experience rather than computation rules learned from supervisions. Here a pretrained StyleGAN is used to capture the practical experience of wearing. Experiments show that projecting the rough alignment of clothing and body onto the StyleGAN space can yield photo-realistic wearing results. Experiments on real scene proprietary model images demonstrate the superiority of DGP over several state-of-the-art supervised methods when generating clothing model images.

cs.CV↗

arXiv2021

Low-Rank Subspaces in GANs

The latent space of a Generative Adversarial Network (GAN) has been shown to encode rich semantics within some subspaces. To identify these subspaces, researchers typically analyze the statistical information from a collection of synthesized data, and the identified subspaces tend to control image attributes globally (i.e., manipulating an attribute causes the change of an entire image). By contrast, this work introduces low-rank subspaces that enable more precise control of GAN generation. Concretely, given an arbitrary image and a region of interest (e.g., eyes of face images), we manage to relate the latent space to the image region with the Jacobian matrix and then use low-rank factorization to discover steerable latent subspaces. There are three distinguishable strengths of our approach that can be aptly called LowRankGAN. First, compared to analytic algorithms in prior work, our low-rank factorization of Jacobians is able to find the low-dimensional representation of attribute manifold, making image editing more precise and controllable. Second, low-rank factorization naturally yields a null space of attributes such that moving the latent code within it only affects the outer region of interest. Therefore, local image editing can be simply achieved by projecting an attribute vector into the null space without relying on a spatial mask as existing methods do. Third, our method can robustly work with a local region from one image for analysis yet well generalize to other images, making it much easy to use in practice. Extensive experiments on state-of-the-art GAN models (including StyleGAN2 and BigGAN) trained on various datasets demonstrate the effectiveness of our LowRankGAN.

cs.CV↗

arXiv2021

MVT: Mask Vision Transformer for Facial Expression Recognition in the wild

Facial Expression Recognition (FER) in the wild is an extremely challenging task in computer vision due to variant backgrounds, low-quality facial images, and the subjectiveness of annotators. These uncertainties make it difficult for neural networks to learn robust features on limited-scale datasets. Moreover, the networks can be easily distributed by the above factors and perform incorrect decisions. Recently, vision transformer (ViT) and data-efficient image transformers (DeiT) present their significant performance in traditional classification tasks. The self-attention mechanism makes transformers obtain a global receptive field in the first layer which dramatically enhances the feature extraction capability. In this work, we first propose a novel pure transformer-based mask vision transformer (MVT) for FER in the wild, which consists of two modules: a transformer-based mask generation network (MGN) to generate a mask that can filter out complex backgrounds and occlusion of face images, and a dynamic relabeling module to rectify incorrect labels in FER datasets in the wild. Extensive experimental results demonstrate that our MVT outperforms state-of-the-art methods on RAF-DB with 88.62%, FERPlus with 89.22%, and AffectNet-7 with 64.57%, respectively, and achieves a comparable result on AffectNet-8 with 61.40%.

cs.CV↗