Kaidong Zhang

Kaidong Zhang, Dong Liu

We propose SAMed, a general solution for medical image segmentation. Different from the previous methods, SAMed is built upon the large-scale image segmentation model, Segment Anything Model (SAM), to explore the new research paradigm of customizing large-scale models for medical image segmentation. SAMed applies the low-rank-based (LoRA) finetuning strategy to the SAM image encoder and finetunes it together with the prompt encoder and the mask decoder on labeled medical image segmentation datasets. We also observe the warmup finetuning strategy and the AdamW optimizer lead SAMed to successful convergence and lower loss. Different from SAM, SAMed could perform semantic segmentation on medical images. Our trained SAMed model achieves 81.88 DSC and 20.64 HD on the Synapse multi-organ segmentation dataset, which is on par with the state-of-the-art methods. We conduct extensive experiments to validate the effectiveness of our design. Since SAMed only updates a small fraction of the SAM parameters, its deployment cost and storage cost are quite marginal in practical usage. The code of SAMed is available at https://github.com/hitachinsk/SAMed.

Kaidong Zhang, Jingjing Fu, Dong Liu

We propose a flow-guided transformer, which innovatively leverage the motion discrepancy exposed by optical flows to instruct the attention retrieval in transformer for high fidelity video inpainting. More specially, we design a novel flow completion network to complete the corrupted flows by exploiting the relevant flow features in a local temporal window. With the completed flows, we propagate the content across video frames, and adopt the flow-guided transformer to synthesize the rest corrupted regions. We decouple transformers along temporal and spatial dimension, so that we can easily integrate the locally relevant completed flows to instruct spatial attention only. Furthermore, we design a flow-reweight module to precisely control the impact of completed flows on each spatial transformer. For the sake of efficiency, we introduce window partition strategy to both spatial and temporal transformers. Especially in spatial transformer, we design a dual perspective spatial MHSA, which integrates the global tokens to the window-based attention. Extensive experiments demonstrate the effectiveness of the proposed method qualitatively and quantitatively. Codes are available at https://github.com/hitachinsk/FGT.

Kaidong Zhang, Ziyang Gan, Dong Liu et al.

Total hip arthroplasty (THA) is a widely used surgical procedure in orthopedics. For THA, it is of clinical significance to analyze the bone structure from the CT images, especially to observe the structure of the acetabulum and femoral head, before the surgical procedure. For such bone structure analyses, deep learning technologies are promising but require high-quality labeled data for the learning, while the data labeling is costly. We address this issue and propose an efficient data annotation pipeline for producing a deep learning-oriented dataset. Our pipeline consists of non-learning-based bone extraction (BE) and acetabulum and femoral head segmentation (AFS) and active-learning-based annotation refinement (AAR). For BE we use the classic graph-cut algorithm. For AFS we propose an improved algorithm, including femoral head boundary localization using first-order and second-order gradient regularization, line-based non-maximum suppression, and anatomy prior-based femoral head extraction. For AAR, we refine the algorithm-produced pseudo labels with the help of trained deep models: we measure the uncertainty based on the disagreement between the original pseudo labels and the deep model predictions, and then find out the samples with the largest uncertainty to ask for manual labeling. Using the proposed pipeline, we construct a large-scale bone structure analyses dataset from more than 300 clinical and diverse CT scans. We perform careful manual labeling for the test set of our data. We then benchmark multiple state-of-the art deep learning-based methods of medical image segmentation using the training and test sets of our data. The extensive experimental results validate the efficacy of the proposed data annotation pipeline. The dataset, related codes and models will be publicly available at https://github.com/hitachinsk/THA.

Kaidong Zhang, Jialun Peng, Jingjing Fu et al.

Transformers have been widely used for video processing owing to the multi-head self attention (MHSA) mechanism. However, the MHSA mechanism encounters an intrinsic difficulty for video inpainting, since the features associated with the corrupted regions are degraded and incur inaccurate self attention. This problem, termed query degradation, may be mitigated by first completing optical flows and then using the flows to guide the self attention, which was verified in our previous work - flow-guided transformer (FGT). We further exploit the flow guidance and propose FGT++ to pursue more effective and efficient video inpainting. First, we design a lightweight flow completion network by using local aggregation and edge loss. Second, to address the query degradation, we propose a flow guidance feature integration module, which uses the motion discrepancy to enhance the features, together with a flow-guided feature propagation module that warps the features according to the flows. Third, we decouple the transformer along the temporal and spatial dimensions, where flows are used to select the tokens through a temporally deformable MHSA mechanism, and global tokens are combined with the inner-window local tokens through a dual perspective MHSA mechanism. FGT++ is experimentally evaluated to be outperforming the existing video inpainting networks qualitatively and quantitatively.

Pengzhen Ren, Kaidong Zhang, Hetao Zheng et al.

Considering how to make the model accurately understand and follow natural language instructions and perform actions consistent with world knowledge is a key challenge in robot manipulation. This mainly includes human fuzzy instruction reasoning and the following of physical knowledge. Therefore, the embodied intelligence agent must have the ability to model world knowledge from training data. However, most existing vision and language robot manipulation methods mainly operate in less realistic simulator and language settings and lack explicit modeling of world knowledge. To bridge this gap, we introduce a novel and simple robot manipulation framework, called Surfer. It is based on the world model, treats robot manipulation as a state transfer of the visual scene, and decouples it into two parts: action and scene. Then, the generalization ability of the model on new instructions and new scenes is enhanced by explicit modeling of the action and scene prediction in multi-modal information. In addition to the framework, we also built a robot manipulation simulator that supports full physics execution based on the MuJoCo physics engine. It can automatically generate demonstration training data and test data, effectively reducing labor costs. To conduct a comprehensive and systematic evaluation of the robot manipulation model in terms of language understanding and physical execution, we also created a robotic manipulation benchmark with progressive reasoning tasks, called SeaWave. It contains 4 levels of progressive reasoning tasks and can provide a standardized testing platform for embedded AI agents in multi-modal environments. On average, Surfer achieved a success rate of 54.74% on the defined four levels of manipulation tasks, exceeding the best baseline performance of 47.64%.

Chang Liu, Shunxin Xu, Jialun Peng et al.

One tough problem of image inpainting is to restore complex structures in the corrupted regions. It motivates interactive image inpainting which leverages additional hints, e.g., sketches, to assist the inpainting process. Sketch is simple and intuitive to end users, but meanwhile has free forms with much randomness. Such randomness may confuse the inpainting models, and incur severe artifacts in completed images. To address this problem, we propose a two-stage image inpainting method termed SketchRefiner. In the first stage, we propose using a cross-correlation loss function to robustly calibrate and refine the user-provided sketches in a coarse-to-fine fashion. In the second stage, we learn to extract informative features from the abstracted sketches in the feature space and modulate the inpainting process. We also propose an algorithm to simulate real sketches automatically and build a test protocol with different applications. Experimental results on public datasets demonstrate that SketchRefiner effectively utilizes sketch information and eliminates the artifacts due to the free-form sketches. Our method consistently outperforms the state-of-the-art ones both qualitatively and quantitatively, meanwhile revealing great potential in real-world applications. Our code and dataset are available.

Yunpeng Qu, Kaidong Zhang, Yukang Ding et al.

Visual generative models based on latent space have achieved great success, underscoring the significance of visual tokenization. Mapping images to latents boosts efficiency and enables multimodal alignment for scaling up in downstream tasks. Existing visual tokenizers primarily map images into fixed 2D spatial grids and focus on pixel-level restoration, which hinders the capture of representations with compact global semantics. To address these issues, we propose \textbf{SemTok}, a semantic one-dimensional tokenizer that compresses 2D images into 1D discrete tokens with high-level semantics. SemTok sets a new state-of-the-art in image reconstruction, achieving superior fidelity with a remarkably compact token representation. This is achieved via a synergistic framework with three key innovations: a 2D-to-1D tokenization scheme, a semantic alignment constraint, and a two-stage generative training strategy. Building on SemTok, we construct a masked autoregressive generation framework, which yields notable improvements in downstream image generation tasks. Experiments confirm the effectiveness of our semantic 1D tokenization. Our code will be open-sourced.

Zitong Huang, Kaidong Zhang, Yukang Ding et al.

Aligning text-to-video diffusion models with human preferences is crucial for generating high-quality videos. Existing Direct Preference Otimization (DPO) methods rely on multi-sample ranking and task-specific critic models, which is inefficient and often yields ambiguous global supervision. To address these limitations, we propose LocalDPO, a novel post-training framework that constructs localized preference pairs from real videos and optimizes alignment at the spatio-temporal region level. We design an automated pipeline to efficiently collect preference pair data that generates preference pairs with a single inference per prompt, eliminating the need for external critic models or manual annotation. Specifically, we treat high-quality real videos as positive samples and generate corresponding negatives by locally corrupting them with random spatio-temporal masks and restoring only the masked regions using the frozen base model. During training, we introduce a region-aware DPO loss that restricts preference learning to corrupted areas for rapid convergence. Experiments on Wan2.1 and CogVideoX demonstrate that LocalDPO consistently improves video fidelity, temporal coherence and human preference scores over other post-training approaches, establishing a more efficient and fine-grained paradigm for video generator alignment.

Kaidong Zhang, Pengzhen Ren, Bingqian Lin et al.

Language-guided robotic manipulation is a challenging task that requires an embodied agent to follow abstract user instructions to accomplish various complex manipulation tasks. Previous work trivially fitting the data without revealing the relation between instruction and low-level executable actions, these models are prone to memorizing the surficial pattern of the data instead of acquiring the transferable knowledge, and thus are fragile to dynamic environment changes. To address this issue, we propose a PrIrmitive-driVen waypOinT-aware world model for Robotic manipulation (PIVOT-R) that focuses solely on the prediction of task-relevant waypoints. Specifically, PIVOT-R consists of a Waypoint-aware World Model (WAWM) and a lightweight action prediction module. The former performs primitive action parsing and primitive-driven waypoint prediction, while the latter focuses on decoding low-level actions. Additionally, we also design an asynchronous hierarchical executor (AHE), which can use different execution frequencies for different modules of the model, thereby helping the model reduce computational redundancy and improve model execution efficiency. Our PIVOT-R outperforms state-of-the-art (SoTA) open-source models on the SeaWave benchmark, achieving an average relative improvement of 19.45% across four levels of instruction tasks. Moreover, compared to the synchronously executed PIVOT-R, the execution efficiency of PIVOT-R with AHE is increased by 28-fold, with only a 2.9% drop in performance. These results provide compelling evidence that our PIVOT-R can significantly improve both the performance and efficiency of robotic manipulation.

Kaidong Zhang, Jian Zhang, Rongtao Xu et al.

Vision--Language--Action (VLA) models have emerged as a powerful paradigm for open-world robot manipulation, but their practical deployment is often constrained by \emph{cost}: billion-scale VLM backbones and iterative diffusion/flow-based action heads incur high latency and compute, making real-time control expensive on commodity hardware. We present A1, a fully open-source and transparent VLA framework designed for low-cost, high-throughput inference without sacrificing manipulation success; Our approach leverages pretrained VLMs that provide implicit affordance priors for action generation. We release the full training stack (training code, data/data-processing pipeline, intermediate checkpoints, and evaluation scripts) to enable end-to-end reproducibility. Beyond optimizing the VLM alone, A1 targets the full inference pipeline by introducing a budget-aware adaptive inference scheme that jointly accelerates the backbone and the \emph{action head}. Specifically, we monitor action consistency across intermediate VLM layers to trigger early termination, and propose Inter-Layer Truncated Flow Matching that warm-starts denoising across layers, enabling accurate actions with substantially fewer effective denoising iterations. Across simulation benchmarks (LIBERO, VLABench) and real robots (Franka, AgiBot), A1 achieves state-of-the-art success rates while significantly reducing inference cost (e.g., up to 72% lower per-episode latency for flow-matching inference and up to 76.6% backbone computation reduction with minor performance degradation). On RoboChallenge, A1 achieves an average success rate of 29.00%, outperforming baselines including pi0(28.33%), X-VLA (21.33%), and RDT-1B (15.00%).

Tianyu Yu, Kechen Fang, Zihao Wan et al.

Most Vision Language Models (VLMs) directly map outputs from ViT encoders to the LLM via a lightweight projector. While effective, recent analysis suggests this architecture suffers from an alignment challenge: visual features remain distant from the text space in the initial layers of the LLM, forcing the model to waste critical depth~\cite{zhang-etal-2024-investigating,artzy-schwartz-2024-attend} on superficial modality alignment rather than deep understanding and complex reasoning. In this work, we propose Deep Pre-Alignment (DPA), a novel architecture that replaces the standard ViT encoder with a small VLM as perceiver, ensuring visual features are deeply aligned with the text space of the target large language model. Comprehensive experiments demonstrate the effectiveness of DPA. On the 4B parameter scale, DPA outperforms baselines by 1.9 points across 8 multimodal benchmarks, with gains widening to 3.0 points at the 32B scale. Moreover, by offloading alignment to the perceiver, DPA achieves a 32.9\% reduction in language capability forgetting over 3 text benchmarks. We further demonstrate that these gains are consistent across different LLM families including Qwen3 and LLaMA 3.2, highlighting the generality of our approach. Beyond performance, DPA also offers a seamless upgrade path for current VLM development, requiring only a modular replacement for the visual encoder with marginal computation overhead.

Weijia Liufu, Xiaoyu Guo, Ruiyi Chen et al.

Vision-Language-Action (VLA) models remain brittle in long-horizon, contact-rich manipulation because success-only imitation provides little supervision for execution drift, while failed rollouts are often discarded. We introduce RePO-VLA, a recovery-driven policy optimization framework that assigns distinct roles to success, recovery, and failure trajectories. RePO-VLA first applies Recovery-Aware Initialization (RAI), slicing recovery segments and resetting history so corrective actions depend on the current adverse state rather than the preceding failure. It then learns a Progress-Aware Semantic Value Function (PAS-VF), aligning spatiotemporal trajectory features with instructions and successful references. The resulting labels salvage useful failure prefixes via reliability decay, while low-value labels mark drift and terminal breakdowns, teaching differences among nominal, failed, and corrective actions. The data engine turns adverse states into planner-generated or human-collected corrective rollouts, teaching recovery to the success manifold. Value-Conditioned Refinement (VCR) trains the policy to prefer high-progress actions. At deployment, a fixed high value ($v=1.0$) biases actions toward the learned success manifold without online failure detectors or heuristic retries. We introduce FRBench, with standardized error injection and recovery-focused evaluation. Across simulated and real-world bimanual tasks, RePO-VLA improves robustness, raising adversarial success from 20% to 75% on average and up to 80% in scaled real-world trials.

Ganlin Yang, Kaidong Zhang, Jingjing Fu et al.

Aliasing artifacts in renderings produced by Neural Radiance Field (NeRF) is a long-standing but complex issue in the field of 3D implicit representation, which arises from a multitude of intricate causes and was mitigated by designing more advanced but complex scene parameterization methods before. In this paper, we present a Diffusion-based restoration method for anti-aliasing Neural Radiance Field (Drantal-NeRF). We consider the anti-aliasing issue from a low-level restoration perspective by viewing aliasing artifacts as a kind of degradation model added to clean ground truths. By leveraging the powerful prior knowledge encapsulated in diffusion model, we could restore the high-realism anti-aliasing renderings conditioned on aliased low-quality counterparts. We further employ a feature-wrapping operation to ensure multi-view restoration consistency and finetune the VAE decoder to better adapt to the scene-specific data distribution. Our proposed method is easy to implement and agnostic to various NeRF backbones. We conduct extensive experiments on challenging large-scale urban scenes as well as unbounded 360-degree scenes and achieve substantial qualitative and quantitative improvements.

Chang Liu, Rui Li, Kaidong Zhang et al.

Recent advancements of generative AI have significantly promoted content creation and editing, where prevailing studies further extend this exciting progress to video editing. In doing so, these studies mainly transfer the inherent motion patterns from the source videos to the edited ones, where results with inferior consistency to user prompts are often observed, due to the lack of particular alignments between the delivered motions and edited contents. To address this limitation, we present a shape-consistent video editing method, namely StableV2V, in this paper. Our method decomposes the entire editing pipeline into several sequential procedures, where it edits the first video frame, then establishes an alignment between the delivered motions and user prompts, and eventually propagates the edited contents to all other frames based on such alignment. Furthermore, we curate a testing benchmark, namely DAVIS-Edit, for a comprehensive evaluation of video editing, considering various types of prompts and difficulties. Experimental results and analyses illustrate the outperforming performance, visual consistency, and inference efficiency of our method compared to existing state-of-the-art studies.

Kaidong Zhang, Rongtao Xu, Pengzhen Ren et al.

Operating robots in open-ended scenarios with diverse tasks is a crucial research and application direction in robotics. While recent progress in natural language processing and large multimodal models has enhanced robots' ability to understand complex instructions, robot manipulation still faces the procedural skill dilemma and the declarative skill dilemma in open environments. Existing methods often compromise cognitive and executive capabilities. To address these challenges, in this paper, we propose RoBridge, a hierarchical intelligent architecture for general robotic manipulation. It consists of a high-level cognitive planner (HCP) based on a large-scale pre-trained vision-language model (VLM), an invariant operable representation (IOR) serving as a symbolic bridge, and a generalist embodied agent (GEA). RoBridge maintains the declarative skill of VLM and unleashes the procedural skill of reinforcement learning, effectively bridging the gap between cognition and execution. RoBridge demonstrates significant performance improvements over existing baselines, achieving a 75% success rate on new tasks and an 83% average success rate in sim-to-real generalization using only five real-world data samples per task. This work represents a significant step towards integrating cognitive reasoning with physical execution in robotic systems, offering a new paradigm for general robotic manipulation.

Menglin Zhang, Xin Luo, Yunwei Lan et al.

Recent advances in NeRF inpainting have leveraged pretrained diffusion models to enhance performance. However, these methods often yield suboptimal results due to their ineffective utilization of 2D diffusion priors. The limitations manifest in two critical aspects: the inadequate capture of geometric information by pretrained diffusion models and the suboptimal guidance provided by existing Score Distillation Sampling (SDS) methods. To address these problems, we introduce GB-NeRF, a novel framework that enhances NeRF inpainting through improved utilization of 2D diffusion priors. Our approach incorporates two key innovations: a fine-tuning strategy that simultaneously learns appearance and geometric priors and a specialized normal distillation loss that integrates these geometric priors into NeRF inpainting. We propose a technique called Balanced Score Distillation (BSD) that surpasses existing methods such as Score Distillation (SDS) and the improved version, Conditional Score Distillation (CSD). BSD offers improved inpainting quality in appearance and geometric aspects. Extensive experiments show that our method provides superior appearance fidelity and geometric consistency compared to existing approaches.

Tao Tang, Likui Zhang, Youpeng Wen et al.

The development of generalist robot manipulation policies has seen significant progress, driven by large-scale demonstration data across diverse environments. However, the high cost and inefficiency of collecting real-world demonstrations hinder the scalability of data acquisition. While existing simulation platforms enable controlled environments for robotic learning, the challenge of bridging the sim-to-real gap remains. To address these challenges, we propose RoboPearls, an editable video simulation framework for robotic manipulation. Built on 3D Gaussian Splatting (3DGS), RoboPearls enables the construction of photo-realistic, view-consistent simulations from demonstration videos, and supports a wide range of simulation operators, including various object manipulations, powered by advanced modules like Incremental Semantic Distillation (ISD) and 3D regularized NNFM Loss (3D-NNFM). Moreover, by incorporating large language models (LLMs), RoboPearls automates the simulation production process in a user-friendly manner through flexible command interpretation and execution. Furthermore, RoboPearls employs a vision-language model (VLM) to analyze robotic learning issues to close the simulation loop for performance enhancement. To demonstrate the effectiveness of RoboPearls, we conduct extensive experiments on multiple datasets and scenes, including RLBench, COLOSSEUM, Ego4D, Open X-Embodiment, and a real-world robot, which demonstrate our satisfactory simulation performance.

Chang Liu, Rui Li, Kaidong Zhang et al.

Diffusion models have demonstrated impressive abilities in generating photo-realistic and creative images. To offer more controllability for the generation process, existing studies, termed as early-constraint methods in this paper, leverage extra conditions and incorporate them into pre-trained diffusion models. Particularly, some of them adopt condition-specific modules to handle conditions separately, where they struggle to generalize across other conditions. Although follow-up studies present unified solutions to solve the generalization problem, they also require extra resources to implement, e.g., additional inputs or parameter optimization, where more flexible and efficient solutions are expected to perform steerable guided image synthesis. In this paper, we present an alternative paradigm, namely Late-Constraint Diffusion (LaCon), to simultaneously integrate various conditions into pre-trained diffusion models. Specifically, LaCon establishes an alignment between the external condition and the internal features of diffusion models, and utilizes the alignment to incorporate the target condition, guiding the sampling process to produce tailored results. Experimental results on COCO dataset illustrate the effectiveness and superior generalization capability of LaCon under various conditions and settings. Ablation studies investigate the functionalities of different components in LaCon, and illustrate its great potential to serve as an efficient solution to offer flexible controllability for diffusion models.