Faithful Where You Look: Anchor-Based Evaluation Hides Text Fabrication in Image-Conditioned Video Generation

Yujia Wang

Home/Archive/arXiv

arXiv · 2611.00079

Faithful Where You Look: Anchor-Based Evaluation Hides Text Fabrication in Image-Conditioned Video Generation

Yujia Wang

View original source ↗Read source document ↗Export metadata ↓

Abstract

Image-conditioned video models are used to show real places, so whether their output stays faithful to the place they were shown is a practical question. The text benchmarks we found that check fidelity against a reference do so at an anchor - an annotated region, typically the sign the task is about. We show that this choice hides most of the failure. We conditioned four models from three vendors on 48 street-level references of Tokyo storefronts and measured text fidelity at the annotated sign and in the periphery, the other text the reference covers. At the anchor, all four models are close to a no-model control (anchor text fidelity 0.90-0.94 against 0.996) and are not reliably separated from one another. In the periphery, 36-54% of generated text regions have no counterpart in the reference, against 1.6% for the control, although every prompt forbade new readable text. Two blinded raters agree on which regions are fabricated (kappa = 0.83-0.96), the result held in preregistered rounds on two vendors whose videos the measure had never seen (42% and 56%), and VBench-I2V's reference-based consistency scores show no evidence of tracking it in our sample. Most fabricated text is legible-looking text where the reference shows none or cannot be read - a form that string-matching metrics discard, which is why our own first measure missed it. A preregistered Latin-script replication in London missed its bar in a first round and met it in an independent second round. We release code, annotations, per-video scores, ratings and every preregistration. Code and data: https://github.com/geoywang/refanchor-bench

Explore related subjects

cs.CV

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Yujia Wang. 2026-10-18. Faithful Where You Look: Anchor-Based Evaluation Hides Text Fabrication in Image-Conditioned Video Generation. https://arxiv.org/abs/2611.00079

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Graph-Aggregated Prototype Learning for Remote Sensing Semantic Change Detection

Semantic change detection (SCD) extends the binary change detection task to provide not only the change locations but also the detailed "from-to" categories in multi-temporal remote sensing data. Such detailed semantic insights into changes offer considerable advantages for a wide array of applications. However, since SCD involves the simultaneous optimization of multiple tasks, the model is prone to negative transfer due to task-specific learning difficulties and conflicting gradient flows. To address this issue, we propose Graph-Aggregated Prototype Learning for Remote Sensing Semantic Change Detection(GAPL-SCD). In this framework, a multi-task joint optimization method is designed to optimize the primary task of semantic segmentation and change detection, along with the auxiliary task of graph aggregation prototype learning. Adaptive weight allocation and gradient rotation methods are used to alleviate the conflict between training tasks and improve multi-task learning capabilities. Specifically, the graph aggregation prototype learning module constructs an interaction graph using high-level features. Prototypes serve as class proxies, enabling category-level domain alignment across time points and reducing interference from irrelevant changes. Additionally, the proposed self-query multi-level feature interaction and bi-temporal feature fusion modules further enhance multi-scale feature representation, improving performance in complex scenes. Experimental results on the SECOND and Landsat-SCD datasets demonstrate that our method achieves state-of-the-art performance, with significant improvements in accuracy and robustness for SCD task.

cs.CV↗

arXiv2026

Compact Attention: Exploiting Structured Spatio-Temporal Sparsity for Fast Video Generation

The quadratic computational complexity of self-attention mechanisms imposes a critical constraint on transformer-based video generation, particularly for synthesizing ultra-long and high-definition sequences. While most sparse attention techniques employ fixed or heuristic patterns, they often struggle to accommodate the heterogeneous and layer-specific redundancies inherent in video diffusion transformers (VDiTs). We find that different attention heads consistently concentrate most of their attention mass in different coarse query-key regions, a behavior we call functional specialization. Although individual attention values vary with the generated content, these regions remain similar across prompts and random seeds. Leveraging these insights, we introduce Compact Attention, a hardware-aware sparse attention framework for efficient long-video synthesis. We quantify this stability and use it to construct searchable, reusable, head-specific masks. Based on the observation, we propose a deformable dual-window sparse pattern mechanism that flexibly approximates local and axial attention regions within one tile-aligned mask family. To avoid manually choosing a mask for each head, we develop a recall-driven offline auto-search algorithm that identifies high-quality sparse masks while leveraging temporal stability for mask reuse. Extensive evaluations on Wan2.1 (14B) and Hunyuan-Video demonstrate that Compact Attention achieves 2.25x/3.03x attention speedups over FlashAttention-2 and 1.20x/1.74x over FlashAttention-3, respectively, while maintaining visual quality and temporal consistency comparable to full-attention baselines. Project Page: https://yo-ava.github.io/Compact-Attention.github.io/

cs.CV↗

arXiv2026

Backpropagation-Free Test-Time Adaptation via Probabilistic Gaussian Alignment

Test-time adaptation (TTA) enhances the zero-shot robustness under distribution shifts by leveraging unlabeled test data during inference. Despite notable advances, several challenges still limit its broader applicability. First, most methods rely on backpropagation or iterative optimization, which limits scalability and hinders real-time deployment. Second, they lack explicit modeling of class-conditional feature distributions. This modeling is crucial for producing reliable decision boundaries and calibrated predictions, but it remains underexplored due to the lack of both source data and supervision at test time. In this paper, we propose ADAPT, an Advanced Distribution-Aware and backPropagation-free Test-time adaptation method. We reframe TTA as a Gaussian probabilistic inference task by modeling class-conditional likelihoods using gradually updated class means and a shared covariance matrix. This enables closed-form, training-free inference. To correct potential likelihood bias, we introduce lightweight regularization guided by CLIP priors and a historical knowledge bank. ADAPT requires no source data, no gradient updates, and no full access to target data, supporting both online and transductive settings. Extensive experiments across diverse benchmarks demonstrate that our method achieves state-of-the-art performance under a wide range of distribution shifts with superior scalability and robustness.

cs.CV↗