arXiv subjects

Zhenyuan Chen

Publications and source records attributed to Zhenyuan Chen.

17 recordsLinked to original sources

arXiv2026

CARE: Confidence-Aware Reasoning for Reliable Medical VQA

Reinforcement Fine-Tuning (RFT) has enabled medical Multimodal Large Language Models (MLLMs) to produce Chain-of-Thought (CoT) reasoning for visual question answering, yet these models suffer from $\textit{confidence miscalibration}$---a systematic gap between expressed certainty and actual diagnostic accuracy that undermines clinical trust. We propose $\textbf{CARE}$, a $\textbf{C}$onfidence-$\textbf{A}$ware medical $\textbf{RE}$asoning framework that jointly optimizes accuracy and calibration through a dual-stage pipeline. First, a scalable Medical-CoT synthesis provides structured cold-start data for Supervised Fine-Tuning. Second, Group Relative Policy Optimization (GRPO) with a novel $\textbf{Confidence-Aware Reward (CAR)}$ mechanism ties the model's confidence to diagnostic correctness within the reward signal. Across three Medical VQA benchmarks, $\textbf{CARE}$ achieves the highest diagnostic accuracy while obtaining the lowest Expected Calibration Error and Hallucination Rate, establishing a foundation for trustworthy clinical decision support. Our code is available at https://github.com/anotherbricki/CARE.

cs.CV↗

arXiv2026

HiFA4: Training-Free 4-bit FlashAttention on Ascend HIF4 NPUs for LLM Inference

We present HiFA4, a post-training operator-level design that executes both QK^T and PV in FlashAttention as 4-bit HIF4 Cube GEMMs for LLM inference on Ascend NPUs, while maintaining the online softmax state in FP16. To our knowledge, HiFA4 is the first Ascend-HIF4-targeted design of this kind evaluated on standard NLP benchmarks. HiFA4 combines two mechanisms. Smooth-QK applies a calibration-static per-channel equivalent rescaling to Q and K after RoPE, transferring quantization difficulty from K to Q without per-tile online reduction at inference. P-Reordering accumulates the softmax normalizer from the same quantized attention weights P_hat used in the PV GEMM, rather than from a higher-precision reconstruction. We show that this inconsistent formulation introduces a coherent output-scaling error, and validate the effect on a Qwen3-8B Layer-0 MMLU trace, where all 3.6M measured attention tiles exhibit net probability-mass loss with median epsilon_bar = -0.064. P-Reordering also allows the normalizer to be fused into the PV Cube GEMM. Across five LLMs, HiFA4 reduces quantization-induced decision drift. On Qwen3-8B, it recovers 37.5% of the accuracy gap introduced by direct HIF4 quantization, narrows the sample-weighted accuracy loss from 1.12 pp to 0.70 pp, reduces BF16-inconsistent MMLU predictions from 16.3% to 8.2%, and cuts MMLU accuracy regressions by 57% (1071 to 465). On Gemma2-9B, mild smoothing keeps HiFA4 within 0.7 pp of BF16 while reducing MMLU regressions by 27%. On LLaMA3.1-8B, Mistral-7B, and Phi-4B, where Smooth-QK is disabled, P-Reordering with the adopted Q-Mean auxiliary still reduces full-set MMLU regressions by 41-52%. A preliminary instruction-scheduling analysis projects a 35.4% critical-path latency reduction relative to BF16 by fusing the softmax normalizer into the PV Cube GEMM; on-hardware validation is left to future work.

cs.LG↗

arXiv2026

Training-free sparse attention based on cumulative energy filtering

Sparse attention accelerates Diffusion Transformers (DiTs) for video generation by computing only the important tokens while skipping the rest. The token selection strategy is key to balancing sparsity and accuracy. We formulate the token filtering process as a dual-goal optimization problem: maximizing sparsity and minimizing accuracy degradation. Existing algorithms cannot fulfill both objectives simultaneously. For example, Top-p only considers the accuracy constraint, while Top-k maintains a fixed computational budget but loosens the accuracy constraint. This paper demonstrates that maintaining a fixed recall rate is sufficient for ensuring accuracy, whereas a fixed threshold is suboptimal for reducing computational cost. Therefore, we propose a dynamic thresholding scheme to improve sparsity while maintaining the same level of accuracy. Furthermore, our algorithm is deeply integrated with Flash Attention (FA), eliminating the need for any additional masking computation overhead. Experimental results on Wan 2.2 validate that, compared to the BLASST algorithm which is also integrated with FA, our dynamic thresholding strategy enhances sparsity from 61.42\% to 82\% with a VBench metric drop of less than 5\%. This results in an approximate 15\% in attention computation and a $1.61\times$ increase in computational efficiency, which is 1.18x higher than that of BLASST.

cs.CV↗

arXiv2026

WOW-Seg: A Word-free Open World Segmentation Model

Open world image segmentation aims to achieve precise segmentation and semantic understanding of targets within images by addressing the infinitely open set of object categories encountered in the real world. However, traditional closed-set segmentation approaches struggle to adapt to complex open world scenarios, while foundation segmentation models such as SAM exhibit notable discrepancies between their strong segmentation capabilities and relatively weaker semantic understanding. To bridge these discrepancies, we propose WOW-Seg, a Word-free Open World Segmentation model for segmenting and recognizing objects from open-set categories. Specifically, WOW-Seg introduces a novel visual prompt module, Mask2Token, which transforms image masks into visual tokens and ensures their alignment with the VLLM feature space. Moreover, we introduce the Cascade Attention Mask to decouple information across different instances. This approach mitigates inter-instance interference, leading to a significant improvement in model performance. We further construct an open world region recognition test benchmark: the Region Recognition Dataset (RR-7K). With 7,662 classes, it represents the most extensive category-rich region recognition dataset to date. WOW-Seg attains strong results on the LVIS dataset, achieving a semantic similarity of 89.7 and a semantic IoU of 82.4. This performance surpasses the previous SOTA while using only one-eighth the parameter count. These results underscore the strong open world generalization capabilities of WOW-Seg. The code and related resources are available at https://github.com/AAwcAA/WOW-Seg-Meta.

cs.CV↗

arXiv2026

SpaceMind++: Toward Allocentric Cognitive Maps for Spatially Grounded Video MLLMs

Recent multimodal large language models (MLLMs) have made remarkable progress in visual understanding and language-based reasoning, yet they lack a persistent world-centered representation for spatially consistent reasoning in 3D environments. Inspired by the mammalian dual-stream system, where semantic and spatial cues are processed separately and integrated into an allocentric cognitive map, we propose SpaceMind++, a video MLLM architecture that explicitly builds a voxelized cognitive map from RGB videos. This map reorganizes fragmented egocentric observations into a shared 3D metric representation, enabling the model to preserve object permanence and spatial topology across changing viewpoints. To make this allocentric representation usable by a pretrained video MLLM without disrupting its native visual-token interface, we introduce Coordinate-Guided Deep Iterative Fusion, a new mechanism that relays map-level spatial knowledge back into the original 2D visual features. This fusion is explicitly guided by coordinate embeddings and 3D Rotary Positional Encoding, which ground semantic interactions in metric 3D space, resembling the entorhinal binding of sensory features to metric space. Extensive experiments show that SpaceMind++ achieves new state-of-the-art performance on VSI-Bench. Furthermore, it demonstrates superior out-of-distribution generalization on SPBench, SITE-Bench, and SPAR-Bench, underscoring its robustness in unseen 3D environments.

cs.CV↗

arXiv2026

EarthBridge: A Solution for 4th Multi-modal Aerial View Image Challenge Translation Track

Cross-modal image-to-image translation among Electro-Optical (EO), Infrared (IR), and Synthetic Aperture Radar (SAR) sensors is essential for comprehensive multi-modal aerial-view analysis. However, translating between these modalities is notoriously difficult due to their distinct electromagnetic signatures and geometric characteristics. This paper presents \textbf{EarthBridge}, a high-fidelity translation framework developed for the 4th Multi-modal Aerial View Image Challenge -- Translation (MAVIC-T). We explore two distinct methodologies: \textbf{Diffusion Bridge Implicit Models (DBIM)}, which we generalize using non-Markovian bridge processes for high-quality deterministic sampling, and \textbf{Contrastive Unpaired Translation (CUT)}, which utilizes contrastive learning for structural consistency. Our EarthBridge framework employs a channel-concatenated UNet denoiser trained with Karras-weighted bridge scalings and a specialized "booting noise" initialization to handle the inherent ambiguity in cross-modal mappings. We evaluate these methods across all four challenge tasks (SAR$\rightarrow$EO, SAR$\rightarrow$RGB, SAR$\rightarrow$IR, RGB$\rightarrow$IR), achieving superior spatial detail and spectral accuracy. Our solution achieved a composite score of 0.38, securing the second position on the MAVIC-T leaderboard. Code is available at https://github.com/Bili-Sakura/EarthBridge-Preview.

cs.CV↗

arXiv2025

RSCC: A Large-Scale Remote Sensing Change Caption Dataset for Disaster Events

Remote sensing is critical for disaster monitoring, yet existing datasets lack temporal image pairs and detailed textual annotations. While single-snapshot imagery dominates current resources, it fails to capture dynamic disaster impacts over time. To address this gap, we introduce the Remote Sensing Change Caption (RSCC) dataset, a large-scale benchmark comprising 62,351 pre-/post-disaster image pairs (spanning earthquakes, floods, wildfires, and more) paired with rich, human-like change captions. By bridging the temporal and semantic divide in remote sensing data, RSCC enables robust training and evaluation of vision-language models for disaster-aware bi-temporal understanding. Our results highlight RSCC's ability to facilitate detailed disaster-related analysis, paving the way for more accurate, interpretable, and scalable vision-language applications in remote sensing. Code and dataset are available at https://github.com/Bili-Sakura/RSCC.

cs.CV↗

arXiv2025

Task-Specific Sparse Feature Masks for Molecular Toxicity Prediction with Chemical Language Models

Reliable in silico molecular toxicity prediction is a cornerstone of modern drug discovery, offering a scalable alternative to experimental screening. However, the black-box nature of state-of-the-art models remains a significant barrier to adoption, as high-stakes safety decisions demand verifiable structural insights alongside predictive performance. To address this, we propose a novel multi-task learning (MTL) framework designed to jointly enhance accuracy and interpretability. Our architecture integrates a shared chemical language model with task-specific attention modules. By imposing an L1 sparsity penalty on these modules, the framework is constrained to focus on a minimal set of salient molecular fragments for each distinct toxicity endpoint. The resulting framework is trained end-to-end and is readily adaptable to various transformer-based backbones. Evaluated on the ClinTox, SIDER, and Tox21 benchmark datasets, our approach consistently outperforms both single-task and standard MTL baselines. Crucially, the sparse attention weights provide chemically intuitive visualizations that reveal the specific fragments influencing predictions, thereby enhancing insight into the model's decision-making process.

cs.CE↗

arXiv2025

Are Image-to-Video Models Good Zero-Shot Image Editors?

Large-scale video diffusion models show strong world simulation and temporal reasoning abilities, but their use as zero-shot image editors remains underexplored. We introduce IF-Edit, a tuning-free framework that repurposes pretrained image-to-video diffusion models for instruction-driven image editing. IF-Edit addresses three key challenges: prompt misalignment, redundant temporal latents, and blurry late-stage frames. It includes (1) a chain-of-thought prompt enhancement module that transforms static editing instructions into temporally grounded reasoning prompts; (2) a temporal latent dropout strategy that compresses frame latents after the expert-switch point, accelerating denoising while preserving semantic and temporal coherence; and (3) a self-consistent post-refinement step that sharpens late-stage frames using a short still-video trajectory. Experiments on four public benchmarks, covering non-rigid editing, physical and temporal reasoning, and general instruction edits, show that IF-Edit performs strongly on reasoning-centric tasks while remaining competitive on general-purpose edits. Our study provides a systematic view of video diffusion models as image editors and highlights a simple recipe for unified video-image generative reasoning.

cs.CV↗

arXiv2025

Representation Entanglement for Generation: Training Diffusion Transformers Is Much Easier Than You Think

REPA and its variants effectively mitigate training challenges in diffusion models by incorporating external visual representations from pretrained models, through alignment between the noisy hidden projections of denoising networks and foundational clean image representations. We argue that the external alignment, which is absent during the entire denoising inference process, falls short of fully harnessing the potential of discriminative representations. In this work, we propose a straightforward method called Representation Entanglement for Generation (REG), which entangles low-level image latents with a single high-level class token from pretrained foundation models for denoising. REG acquires the capability to produce coherent image-class pairs directly from pure noise, substantially improving both generation quality and training efficiency. This is accomplished with negligible additional inference overhead, requiring only one single additional token for denoising (<0.5\% increase in FLOPs and latency). The inference process concurrently reconstructs both image latents and their corresponding global semantics, where the acquired semantic knowledge actively guides and enhances the image generation process. On ImageNet 256$\times$256, SiT-XL/2 + REG demonstrates remarkable convergence acceleration, achieving $\textbf{63}\times$ and $\textbf{23}\times$ faster training than SiT-XL/2 and SiT-XL/2 + REPA, respectively. More impressively, SiT-L/2 + REG trained for merely 400K iterations outperforms SiT-XL/2 + REPA trained for 4M iterations ($\textbf{10}\times$ longer). Code is available at: https://github.com/Martinser/REG.

cs.CV↗

arXiv2025

NeuralDB: Scaling Knowledge Editing in LLMs to 100,000 Facts with Neural KV Database

Efficiently editing knowledge stored in large language models (LLMs) enables model updates without large-scale training. One possible solution is Locate-and-Edit (L\&E), allowing simultaneous modifications of a massive number of facts. However, such editing may compromise the general abilities of LLMs and even result in forgetting edited facts when scaling up to thousands of edits. In this paper, we model existing linear L\&E methods as querying a Key-Value (KV) database. From this perspective, we then propose NeuralDB, an editing framework that explicitly represents the edited facts as a neural KV database equipped with a non-linear gated retrieval module, % In particular, our gated module only operates when inference involves the edited facts, effectively preserving the general abilities of LLMs. Comprehensive experiments involving the editing of 10,000 facts were conducted on the ZsRE and CounterFacts datasets, using GPT2-XL, GPT-J (6B) and Llama-3 (8B). The results demonstrate that NeuralDB not only excels in editing efficacy, generalization, specificity, fluency, and consistency, but also preserves overall performance across six representative text understanding and generation tasks. Further experiments indicate that NeuralDB maintains its effectiveness even when scaled to 100,000 facts (\textbf{50x} more than in prior work).

cs.CL↗

arXiv2025

Rethinking Inductive Bias in Geographically Neural Network Weighted Regression

Inductive bias is a key factor in spatial regression models, determining how well a model can learn from limited data and capture spatial patterns. This work revisits the inductive biases in Geographically Neural Network Weighted Regression (GNNWR) and identifies limitations in current approaches for modeling spatial non-stationarity. While GNNWR extends traditional Geographically Weighted Regression by using neural networks to learn spatial weighting functions, existing implementations are often restricted by fixed distance-based schemes and limited inductive bias. We propose to generalize GNNWR by incorporating concepts from convolutional neural networks, recurrent neural networks, and transformers, introducing local receptive fields, sequential context, and self-attention into spatial regression. Through extensive benchmarking on synthetic spatial datasets with varying heterogeneity, noise, and sample sizes, we show that GNNWR outperforms classic methods in capturing nonlinear and complex spatial relationships. Our results also reveal that model performance depends strongly on data characteristics, with local models excelling in highly heterogeneous or small-sample scenarios, and global models performing better with larger, more homogeneous data. These findings highlight the importance of inductive bias in spatial modeling and suggest future directions, including learnable spatial weighting functions, hybrid neural architectures, and improved interpretability for models handling non-stationary spatial data.

cs.LG↗

arXiv2025

HazyDet: Open-Source Benchmark for Drone-View Object Detection with Depth-Cues in Hazy Scenes

Object detection from aerial platforms under adverse atmospheric conditions, particularly haze, is paramount for robust drone autonomy. Yet, this domain remains largely underexplored, primarily hindered by the absence of specialized benchmarks. To bridge this gap, we present \textit{HazyDet}, the first, large-scale benchmark specifically designed for drone-view object detection in hazy conditions. Comprising 383,000 real-world instances derived from both naturally hazy captures and synthetically hazed scenes augmented from clear images, HazyDet provides a challenging and realistic testbed for advancing detection algorithms. To address the severe visual degradation induced by haze, we propose the Depth-Conditioned Detector (DeCoDet), a novel architecture that integrates a Depth-Conditioned Kernel to dynamically modulate feature representations based on depth cues. The practical efficacy and robustness of DeCoDet are further enhanced by its training with a Progressive Domain Fine-Tuning (PDFT) strategy to navigate synthetic-to-real domain shifts, and a Scale-Invariant Refurbishment Loss (SIRLoss) to ensure resilient learning from potentially noisy depth annotations. Comprehensive empirical validation on HazyDet substantiates the superiority of our unified DeCoDet framework, which achieves state-of-the-art performance, surpassing the closest competitor by a notable +1.5\% mAP on challenging real-world hazy test scenarios. Our dataset and toolkit are available at https://github.com/GrokCV/HazyDet.

cs.CV↗

arXiv2024

Agent-based Video Trimming

As information becomes more accessible, user-generated videos are increasing in length, placing a burden on viewers to sift through vast content for valuable insights. This trend underscores the need for an algorithm to extract key video information efficiently. Despite significant advancements in highlight detection, moment retrieval, and video summarization, current approaches primarily focus on selecting specific time intervals, often overlooking the relevance between segments and the potential for segment arranging. In this paper, we introduce a novel task called Video Trimming (VT), which focuses on detecting wasted footage, selecting valuable segments, and composing them into a final video with a coherent story. To address this task, we propose Agent-based Video Trimming (AVT), structured into three phases: Video Structuring, Clip Filtering, and Story Composition. Specifically, we employ a Video Captioning Agent to convert video slices into structured textual descriptions, a Filtering Module to dynamically discard low-quality footage based on the structured information of each clip, and a Video Arrangement Agent to select and compile valid clips into a coherent final narrative. For evaluation, we develop a Video Evaluation Agent to assess trimmed videos, conducting assessments in parallel with human evaluations. Additionally, we curate a new benchmark dataset for video trimming using raw user videos from the internet. As a result, AVT received more favorable evaluations in user studies and demonstrated superior mAP and precision on the YouTube Highlights, TVSum, and our own dataset for the highlight detection task. The code and models are available at https://ylingfeng.github.io/AVT.

cs.CV↗

arXiv2024

Revisiting Prompt Pretraining of Vision-Language Models

Prompt learning is an effective method to customize Vision-Language Models (VLMs) for various downstream tasks, involving tuning very few parameters of input prompt tokens. Recently, prompt pretraining in large-scale dataset (e.g., ImageNet-21K) has played a crucial role in prompt learning for universal visual discrimination. However, we revisit and observe that the limited learnable prompts could face underfitting risks given the extensive images during prompt pretraining, simultaneously leading to poor generalization. To address the above issues, in this paper, we propose a general framework termed Revisiting Prompt Pretraining (RPP), which targets at improving the fitting and generalization ability from two aspects: prompt structure and prompt supervision. For prompt structure, we break the restriction in common practice where query, key, and value vectors are derived from the shared learnable prompt token. Instead, we introduce unshared individual query, key, and value learnable prompts, thereby enhancing the model's fitting capacity through increased parameter diversity. For prompt supervision, we additionally utilize soft labels derived from zero-shot probability predictions provided by a pretrained Contrastive Language Image Pretraining (CLIP) teacher model. These soft labels yield more nuanced and general insights into the inter-class relationships, thereby endowing the pretraining process with better generalization ability. RPP produces a more resilient prompt initialization, enhancing its robust transferability across diverse visual recognition tasks. Experiments across various benchmarks consistently confirm the state-of-the-art (SOTA) performance of our pretrained prompts. Codes and models will be made available soon.

cs.CV↗

arXiv2022

ORBGRAND Is Almost Capacity-Achieving

Decoding via sequentially guessing the error pattern in a received noisy sequence has received attention recently, and ORBGRAND has been proposed as one such decoding algorithm that is capable of utilizing the soft information embedded in the received noisy sequence. An information theoretic study is conducted for ORBGRAND, and it is shown that the achievable rate of ORBGRAND using independent and identically distributed random codebooks almost coincides with the channel capacity, for an additive white Gaussian noise channel under antipodal input. For finite-length codes, improved guessing schemes motivated by the information theoretic study are proposed that attain lower error rates than ORBGRAND, especially in the high signal-to-noise ratio regime.

cs.IT↗

arXiv2020

LID 2020: The Learning from Imperfect Data Challenge Results

Learning from imperfect data becomes an issue in many industrial applications after the research community has made profound progress in supervised learning from perfectly annotated datasets. The purpose of the Learning from Imperfect Data (LID) workshop is to inspire and facilitate the research in developing novel approaches that would harness the imperfect data and improve the data-efficiency during training. A massive amount of user-generated data nowadays available on multiple internet services. How to leverage those and improve the machine learning models is a high impact problem. We organize the challenges in conjunction with the workshop. The goal of these challenges is to find the state-of-the-art approaches in the weakly supervised learning setting for object detection, semantic segmentation, and scene parsing. There are three tracks in the challenge, i.e., weakly supervised semantic segmentation (Track 1), weakly supervised scene parsing (Track 2), and weakly supervised object localization (Track 3). In Track 1, based on ILSVRC DET, we provide pixel-level annotations of 15K images from 200 categories for evaluation. In Track 2, we provide point-based annotations for the training set of ADE20K. In Track 3, based on ILSVRC CLS-LOC, we provide pixel-level annotations of 44,271 images for evaluation. Besides, we further introduce a new evaluation metric proposed by \cite{zhang2020rethinking}, i.e., IoU curve, to measure the quality of the generated object localization maps. This technical report summarizes the highlights from the challenge. The challenge submission server and the leaderboard will continue to open for the researchers who are interested in it. More details regarding the challenge and the benchmarks are available at https://lidchallenge.github.io

cs.CV↗