Searcharxiv⌕ Search

arXiv subjects

Genpei Zhang

Publications and source records attributed to Genpei Zhang.

11 recordsLinked to original sources

OccJEPA: Occlusion as Masking for Amodal Video Segmentation

Amodal video segmentation asks for the complete mask of an object in every frame, including the parts hidden behind other objects. The best current methods fine-tune video diffusion models to generate these masks; they need a depth network and tens of denoising steps, and their outputs vary from sample to sample. Self-supervised video world models of the JEPA family are pre-trained on a closely related task, predicting the representations of masked space-time regions from the visible ones, yet their predictor is usually discarded after pre-training. We reuse it by treating occlusion as masking. Tokens not covered by the object's visible mask are dropped, the V-JEPA 2.1 predictor is fine-tuned to fill them in, and a lightweight decoder turns all features into amodal masks. The resulting model, OccJEPA (Occlusion-as-masking JEPA), is deterministic and 32x faster per frame than Diffusion-VAS, the diffusion-based state of the art. It is also more accurate on SAIL-VOS (+3.70 occluded-region mIoU over all occluded frames), on TAO-Amodal (+4.00 AP50 on a held-out test split, +4.32 with SAM2-tracked masks) and on Kubric. Controlled experiments trace most of the gain to the predictor's pre-trained weights rather than its capacity, and the predictor helps at every encoder size tested (87M to 1.85B parameters).

cs.CV↗

MissPath-FM: Flow Matching with Structured Priors for Partially Observed Time Series

Flow matching has emerged as a powerful alternative to diffusion models for generative modeling, but standard conditional flow matching is fundamentally misaligned with partially observed time series: starting from an uninformed Gaussian source forces the model to transport even already-observed entries from noise to the target, wasting capacity on trivially known regions. We propose MissPath-FM, a partial-observation-aware flow matching framework that redesigns the source prior and probability path for incomplete time series. MissPath-FM employs structured source priors (interpolation priors and Gaussian process posterior priors) to initialize the flow close to the observed signal, paired with a missingness-aware smooth path that allocates transport asymmetrically between observed and missing regions. We evaluate systematically across four datasets, four missingness mechanisms (MCAR, block, MNAR, irregular), two missing ratios, and three sequence lengths (T in {24, 48, 96}), totaling 48 settings against eight baselines. MissPath-FM wins 29 of 32 settings at T=48 and 45 of 48 overall (93.8%). Our ablations reveal a regime-dependent design hierarchy in which no single component suffices: under random missingness the source prior yields a large initial gain, but under structured block missingness it can fail entirely and the learned flow becomes indispensable; the missingness-aware path contributes a further 12%-26% across all regimes. Mechanistically, structured priors shorten observed-region transport by up to 15x, concentrating the velocity field on genuinely uncertain entries. Project page: https://zgp110.github.io/misspath-fm/

cs.LG↗

Linear Multi-granular Body Modeling for Text-Driven Motion Generation

Text-to-motion generation is a rapidly growing field at the nexus of multimodal learning and computer graphics, enabling flexible and cost-effective applications in gaming, animation, robotics, and virtual reality. Existing part-aware pipelines typically split the body into segments and then recover whole-body coordination via attention-based fusion, which incurs quadratic complexity and fails to model part-level and whole-body cues simultaneously. Moreover, we observe that naively stacking spatial and temporal modules accumulates redundant features, thereby requiring deeper networks to achieve competitive fidelity. To address these issues, we propose HiSTF Mamba, a linear-time framework composed of Dual-Spatial Mamba, Bi-Temporal Mamba, and the Dynamic Spatiotemporal Fusion Module (DSFM). Dual-Spatial Mamba performs parallel part-based and whole-based modeling to simultaneously capture local joint dynamics and global coordination. Bi-Temporal Mamba employs bidirectional scanning to encode both short-term motion details and long-term dependencies. DSFM further removes redundancy and extracts complementary temporal information before fusing it with spatial features, yielding an expressive spatiotemporal representation. Extensive experiments on HumanML3D and KIT-ML demonstrate that HiSTF Mamba improves generation fidelity and semantic alignment while requiring substantially fewer parameters than existing segment-based architectures.

cs.CV↗

Can Scene Text Recognition Read Rare Compositions?

Scene text recognition is reported as 89--97% accurate on the six standard benchmarks, and the problem is widely treated as saturated. We present an alternative reading. When the same test images are stratified jointly by ground-truth word rarity and character n-gram novelty against a reference corpus, accuracy at the rare-word x rare-trigram corner of the resulting 5x5 grid drops 10--18 pt below the q3/q3 centre across nine English specialised recognisers, and the same direction (corner below centre) holds on all 13 of 13 (language, model) pairs we test across four writing systems (Latin, Han, Han+kana, Arabic). The drop is not a capacity bottleneck. A 6x vision-backbone scale-up (CLIP4STR-Base 158M -> CLIP4STR-Huge 1.0B, OpenCLIP ViT-H/14 LAION-2B) leads every benchmark in aggregate accuracy yet leaves the stress corner unchanged (86.9 -> 86.5, within paired-bootstrap noise). Four converging probes--layer-wise probing, confidence-when-wrong, attention re-balancing, and a cross-script commit-vs-abstain error split--localise the failure to the autoregressive decoder's lexical prior. We then ask how much of the gap existing techniques recover. Of 16 non-architectural mitigations, the largest mean q5/q5 gain is +1.3 pt and none clears the paired-bootstrap noise floor; the only intervention that does is the architectural shift from autoregressive to CTC decoding (SVTRv2, +2.5 pt, p=0.02, n=474). A confidence-routed AR-CTC ensemble adds a directionally consistent +0.6 pt that stays within noise, and its dominant learned coefficient is each model's own minimum-softmax confidence--independently echoing the mechanism above. No configuration we test improves both the compositional corner and aggregate accuracy. The rare-input long tail thus points to architectural change rather than added capacity.

cs.CV↗

The Visual Insensitivity Gap: Diagnosing When Vision-Language Models Fail to Use Visual Evidence

Vision-language models are evaluated by aggregate accuracy on multimodal benchmarks, a practice that implicitly assumes the model uses its visual input. We show this assumption fails on 40%--97% of samples across six VLMs and three perceptual benchmarks: blurring the question-relevant visual region leaves the next-token distribution nearly unchanged. We name this phenomenon the Visual Insensitivity Gap and quantify it with a per-sample Visual Sensitivity Index (VSI). The gap is a property of samples, not of models: VSI ranks correlate across models (grand-mean Spearman rho=+0.40, permutation p<10^-3), so the same samples are flagged insensitive by VLMs sharing no architectural detail beyond a contrastively pretrained vision tower. The mechanism is concrete: on the insensitive samples, a linear probe on each model's own vision tower distinguishes perturbed from clean images at 0.72--0.79 accuracy, yet the model's argmax token changes on only 2%--11% of the same samples, an encoder--LLM gap above 0.65 on every model. Mapping VSI's diagnostic utility cell by cell surfaces a strong regime (multi-choice reasoning on capable VLMs: AUROC=0.85--0.87) and a weak regime (well-calibrated factuality, where softmax confidence already leads). VSI is not a universal best abstention signal; it is a sample-intrinsic indicator of vision-ignoring failure, best used as a conditional ensemble component.

cs.CV↗

Encoded but Disconnected: Decomposing Vision-Language Model Failures under a Patching Null

Across three vision-language model architectures (LLaVA-1.5-7B, Qwen2.5-VL-7B, InternVL3-8B), we report a universal negative finding for mid-layer interpretability. On POPE -- the benchmark common to all three -- the mid layers encode the ground-truth answer in 68-91% of errors, yet this signal is not causally active for the final prediction: residual-stream patching yields 0% non-trivial flip at the layer level on all three architectures, and on two of three at the per-head level (Qwen: 0/12,600 patched forwards). The lone exception, InternVL3 layer-20 head-2, is a non-vocab, self-attending head whose effect is localized to that specific head (p < 1e-4). Despite the null, the errors separate operationally into three failure modes -- Perception Failure, Encoded-but-Disconnected, Prior-Override -- learnable above 60% on all three architectures, and the architecture's prior direction predicts which of two interventions elicits a category-specific response. We report these mitigation effects under oracle labels as evidence the categories are mechanistically real, not as a deployable method.

cs.CV↗

Symbolic Augmentation Closes a Canonical-Equivalence Blind Spot in Neural Fact-Checkers

Large language models hallucinate numbers and units when summarizing scientific text, a failure mode that can silently invert a scientific claim. We recast the detection of such errors as typed verification: we introduce a five-class typed-quantity error taxonomy and a 1500-item benchmark, rewritten from PMC and arXiv sources and labeled by two independent LLM annotators with adjudication (Krippendorff's alpha = 0.882). A ModernBERT encoder fine-tuned on this benchmark reaches macro-F1 = 0.899, far above any off-the-shelf neural fact-checker, yet four probes expose a sharp structural blind spot: on canonical-equivalent rewrites of physically equivalent quantities (e.g., 95°C and 368.15 K) its accuracy collapses to 36.5%. We propose Symbolic Augmentation, a training-time framework that runs the modules of a symbolic verifier in reverse to generate label-preserving augmented training data. The augmentation lifts canonical-equivalence robustness to 98.2% while slightly improving in-distribution accuracy (macro-F1: 0.899 to 0.902); the augmented encoder matches a closed-frontier LLM at no inference cost and transfers to an external benchmark (SciFact-Open binary macro-F1: 0.791 to 0.828). Two negative results sharpen the claim: symbolic features as auxiliary encoder inputs add nothing, and symbolic silver labels scale negatively under teacher noise. Together these results identify training-time augmentation as the right integration point between symbolic and learned components.

cs.AI↗

CryoCCD: Conditional Cycle-consistent Diffusion with Biophysical Modeling for Cryo-EM Synthesis

Single-particle cryo-electron microscopy (cryo-EM) has become a cornerstone of structural biology, enabling near-atomic resolution analysis of macromolecules through advanced computational methods. However, the development of cryo-EM processing tools is constrained by the scarcity of high-quality annotated datasets. Synthetic data generation offers a promising alternative, but existing approaches lack thorough biophysical modeling of heterogeneity and fail to reproduce the complex noise observed in real imaging. To address these limitations, we present CryoCCD, a synthesis framework that unifies versatile biophysical modeling with the first conditional cycle-consistent diffusion model tailored for cryo-EM. The biophysical engine provides multi-functional generation capabilities to capture authentic biological organization, and the diffusion model is enhanced with cycle consistency and mask-guided contrastive learning to ensure realistic noise while preserving structural fidelity. Extensive experiments demonstrate that CryoCCD generates structurally faithful micrographs, enhances particle picking and pose estimation, as well as achieves superior performance over state-of-the-art baselines, while also generalizing effectively to held-out protein families.

cs.CV↗

Towards Foundation Models for Cryo-ET Subtomogram Analysis

Cryo-electron tomography (cryo-ET) enables in situ visualization of macromolecular structures, where subtomogram analysis tasks such as classification, alignment, and averaging are critical for structural determination. However, effective analysis is hindered by scarce annotations, severe noise, and poor generalization. To address these challenges, we take the first step towards foundation models for cryo-ET subtomograms. First, we introduce CryoEngine, a large-scale synthetic data generator that produces over 904k subtomograms from 452 particle classes for pretraining. Second, we design an Adaptive Phase Tokenization-enhanced Vision Transformer (APT-ViT), which incorporates adaptive phase tokenization as an equivariance-enhancing module that improves robustness to both geometric and semantic variations. Third, we introduce a Noise-Resilient Contrastive Learning (NRCL) strategy to stabilize representation learning under severe noise conditions. Evaluations across 24 synthetic and real datasets demonstrate state-of-the-art (SOTA) performance on all three major subtomogram tasks and strong generalization to unseen datasets, advancing scalable and robust subtomogram analysis in cryo-ET.

cs.CV↗

EgoPrivacy: What Your First-Person Camera Says About You?

While the rapid proliferation of wearable cameras has raised significant concerns about egocentric video privacy, prior work has largely overlooked the unique privacy threats posed to the camera wearer. This work investigates the core question: How much privacy information about the camera wearer can be inferred from their first-person view videos? We introduce EgoPrivacy, the first large-scale benchmark for the comprehensive evaluation of privacy risks in egocentric vision. EgoPrivacy covers three types of privacy (demographic, individual, and situational), defining seven tasks that aim to recover private information ranging from fine-grained (e.g., wearer's identity) to coarse-grained (e.g., age group). To further emphasize the privacy threats inherent to egocentric vision, we propose Retrieval-Augmented Attack, a novel attack strategy that leverages ego-to-exo retrieval from an external pool of exocentric videos to boost the effectiveness of demographic privacy attacks. An extensive comparison of the different attacks possible under all threat models is presented, showing that private information of the wearer is highly susceptible to leakage. For instance, our findings indicate that foundation models can effectively compromise wearer privacy even in zero-shot settings by recovering attributes such as identity, scene, gender, and race with 70-80% accuracy. Our code and data are available at https://github.com/williamium3000/ego-privacy.

cs.CV↗

PolyhedronNet: Representation Learning for Polyhedra with Surface-attributed Graph

Ubiquitous geometric objects can be precisely and efficiently represented as polyhedra. The transformation of a polyhedron into a vector, known as polyhedra representation learning, is crucial for manipulating these shapes with mathematical and statistical tools for tasks like classification, clustering, and generation. Recent years have witnessed significant strides in this domain, yet most efforts focus on the vertex sequence of a polyhedron, neglecting the complex surface modeling crucial in real-world polyhedral objects. This study proposes \textbf{PolyhedronNet}, a general framework tailored for learning representations of 3D polyhedral objects. We propose the concept of the surface-attributed graph to seamlessly model the vertices, edges, faces, and their geometric interrelationships within a polyhedron. To effectively learn the representation of the entire surface-attributed graph, we first propose to break it down into local rigid representations to effectively learn each local region's relative positions against the remaining regions without geometric information loss. Subsequently, we propose PolyhedronGNN to hierarchically aggregate the local rigid representation via intra-face and inter-face geometric message passing modules, to obtain a global representation that minimizes information loss while maintaining rotation and translation invariance. Our experimental evaluations on four distinct datasets, encompassing both classification and retrieval tasks, substantiate PolyhedronNet's efficacy in capturing comprehensive and informative representations of 3D polyhedral objects. Code and data are available at {https://github.com/dyu62/3D_polyhedron}.

cs.CV↗