Long-horizon video generation is evaluated with whole-frame metrics that reward motion and temporal consistency. For fixed-camera nature scenes this creates an ambiguity: motion of water, fire, smoke, or rain is desirable, whereas motion of the background is an error. A system can therefore score well on motion while its scene drifts, or on consistency while its flow stagnates. We introduce SNF-Bench, an evaluation framework for long-horizon fixed-camera generation that partitions each scene into static support and dynamic flow and reports static fidelity, flow persistence with absolute magnitude, and drift leakage separately, never as one score. Drift leakage is interpretive context rather than a headline measurement. Each factor is validated mechanistically rather than by correlation with preference: we inject global translation, rotation, and scale drift and progressive late freezing at known severity into real generations, and require each factor to respond in its stated direction and to remain selective against corruptions it does not target. Auditing publicly released long-horizon text-conditioned checkpoints under one recorded common inference configuration, plus an image-conditioned track with released-pipeline references and a deployment-sensitivity panel, we find that whole-frame motion and static-region drift induce near-opposite orderings of the same outputs. At maximum controlled translation, fBD and NBF rise to $1.86\times$ and $1.32\times$ baseline, but whole-frame Dynamic Degree reaches only $1.07\times$---rewarding the corruption. SNF-Bench measures where motion occurs and whether it persists; it does not measure physical realism. Project page: https://minar09.github.io/snfbench/.
Recent open-vocabulary segmentation models have advanced semantic perception for UAVs, but predictions from moving aerial platforms can remain temporally inconsistent across repeated observations of the same physical scene. We investigate temporal semantic stability by associating frame-wise predictions with persistent world-space locations through metric 3D fusion. We introduce a voxel-level evaluation framework that jointly characterises final semantic agreement, Semantic Belief Drift (SBD), Observation Persistence (OP), and semantic uncertainty. Experiments on UAVid-3D reveal substantial frame-wise semantic flicker and show that high aggregate world-space agreement can overstate temporal stability when locations have limited repeated-observation support. Persistence-stratified analysis shows that recurrent voxels expose greater semantic disagreement, while belief drift decreases as additional evidence accumulates. This behaviour is observed across two segmentation backbones and remains consistent under variations in voxel resolution, geometric association, and temporal sampling density. Conditions that reduce world-space recurrence can increase apparent aggregate stability, demonstrating that semantic consistency must be interpreted together with observation support. Our findings highlight observation persistence as an essential conditioning variable for evaluating long-horizon semantic reliability.
Andrii Kompanets, Finn Michael Sherry, Remco Duits +2cs.CV
Bridge structures are regularly inspected for structural damage such as cracks and corrosion in order to ensure public safety and reduce maintenance costs. Much research has been done on automating this process using computer vision methods, which are often evaluated and compared using metrics such as intersection over union, mean average precision, etc. However, predicting the actual effectiveness of an inspection method within the field of structural engineering from these metrics remains challenging. To enable the systematic use of these increasingly popular methods in engineering practice, evaluating the performance of these methods in a way that is compatible with standard engineering approaches is therefore an urgent necessity. We present a new statistical evaluation framework to allow the comparison of computer vision methods with conventional visual inspection for crack detection in steel bridges. The framework is based on probability of detection curves and can account for the influence of image resolution. We apply this evaluation method to the real-world ``Cracks in Steel Bridges'' dataset, which contains annotated images of cracks in bridge structures. The quantification of the probability of detection and its uncertainty enables a practical assessment of the effect of automated methods for damage detection in structural reliability analyses. In turn, this enables the wide-spread use of automated (AI-based) damage detection in safety critical applications. This evaluation method provides evidence that the proposed computer vision approach approach is robust for the crack detection task and can have a high added value as an addition to conventional visual inspection methods.
Recognizing human behavior across levels of abstraction, from atomic actions to long-horizon intentions, requires data annotated along a semantic hierarchy. Large corpora provide isolated, atomically labeled clips without temporal composition, whereas recorded composite-activity corpora offer shallow, domain-narrow, fixedhierarchies. A benchmark-generation and evaluation frameworkis proposed that synthesizes a four-level hierarchical-intention benchmark, spanning actions, activities, low-level intentions (LLIs), and high-level intentions (HLIs), from a flat single-label action corpus while retaining real pre-extracted features at the action level. Episodes are assembled by a transition model under a subject-consistency constraint, and a coverage-aware sampler reduces the subject usage Gini from 0.566 to 0.248. Synthesizing such a benchmark raises a circular-supervision risk that recorded datasets avoid: if the rules generating the episodes also govern the evaluation, models can succeed by recovering the generator rather than through genuine reasoning. Validity is addressed by design, holding sequence-generation rules disjoint from the first-order-logic rules used at evaluation. The instantiation yields 15,002 episodes. Four reference baselines from different model families characterize difficulty, not as recognition methods. A compositional held-out gap of 0.13 to 0.17 macro-F1 appears across all baselines, including a graph-aware model that recognizes best yet does not close the gap, indicating a structural property of the benchmark rather than a model artifact. A logic-free baseline still violates the held-out semantic rules above their intrinsic data rate, and the order-destroying control changes macro-F1 within seed variation, serving as a generator-consistency check. Theontology, transition model, and generator are released so the benchmark can beregenerated and extended.
Spatial intelligence is essential for agents to move from static semantic understanding toward interacting with the physical world. Many spatial tasks are grounded in continuous visual scenes, where locations, regions, and paths are more naturally expressed by pointing, marking, or drawing than by reporting precise coordinates or discrete textual symbols. Yet existing spatial reasoning benchmarks usually require coordinates, options, or text, creating an answer-interface mismatch for image-generation models. This makes it difficult to evaluate image-generation models under the same task semantics as text-output VLMs, despite their ability to externalize spatial judgments directly in pixel space. We propose ProVisE (Protocolized Visual Evaluation), a benchmark-agnostic framework that elicits protocol-constrained visual answers from image-generation models and parses them into structured predictions compatible with original metrics. ProVisE also includes an Agentic builder that constructs and validates task-specific protocols for new benchmarks. We further introduce SpatialGen-Bench, a curated diagnostic benchmark of 470 samples across 14 spatial subtasks, four capability levels, and diverse answer forms. We evaluate representative text-output VLMs and image-generation models in a unified setting and validate Agentic protocol construction on six external spatial benchmarks. Results show that image-generation models are competitive when spatial answers can be externalized directly in pixel space, while text-output VLMs retain a clear advantage in compositional spatial reasoning. These findings reveal complementary strengths of pixel-space expression and text-based reasoning and establish a metric-compatible testbed for studying spatial cognition in image-generation models.
Text-to-video (T2V) generators can synthesize realistic and temporally coherent videos, but controllably removing a target concept from a generator remains difficult. Unlike text-to-image concept erasure, T2V unlearning must suppress a target concept that may persist across frames while preserving non-target subjects, actions, scenes, and temporal structure. We propose \textbf{SIRUS}, a training-free inference-time framework for concept-level T2V unlearning. Given textual aliases of a target concept, SIRUS localizes target-related prompt evidence and suppresses target expression during sampling, without updating the text encoder or denoising network. We further introduce a video-oriented evaluation framework for T2V unlearning that separately measures target forgetting, non-target preservation, video quality, jailbreak robustness, and efficiency, using video-level failure criteria, frame-level residue statistics, paired preservation analysis, VBench-based quality diagnostics, and deployment overhead measurement. Across five safety, object, and style concepts on CogVideoX, SIRUS reaches 70.4\% average forgetting success and 25.7\% average frame hit, compared with 44.4\% / 47.2\% for VideoEraser, while reducing the average VBench quality drop from -0.043 to -0.016, yielding the strongest forgetting-quality trade-off among fully evaluated baselines. Transfer experiments on Wan2.2 further suggest that SIRUS generalizes across modern T2V backbones.
Do vision models see colors the way humans do? Existing evaluations of color representations usually compare them with geometric spaces such as CIELAB or with discrete color labels. These references capture perceptual distance or category membership, but not the graded way in which people organize colors. We evaluate color grounding against a fuzzy perceptual model with 86 graded categories fitted to human survey data. The framework can be applied to any image encoder and measures three complementary properties: category boundaries, category compactness, and graded alignment beyond what color geometry alone can explain. Across eleven Vision Transformer encoders, the category-level results are broadly similar, whereas graded alignment differs substantially. Masked Autoencoders achieve the strongest beyond-geometry alignment, with confidence intervals that do not overlap those of the other encoders. A layer-wise analysis further shows that masked reconstruction preserves this structure toward the output. On natural images, MAE represents surface color globally, while language-supervised models encode color more strongly in relation to the foreground object. These results show that human-like color grounding has several distinct aspects that should not be reduced to a single score.
Amir Pouladi, Vesal Ahsani, Haijun Li +2cs.CV cs.LG eess.IV
Fog severely degrades the visibility of small unmanned aerial vehicles (UAVs) in skydominant, long-range imagery, reducing the reliability of downstream detection and tracking. This paper presents a task-driven evaluation framework that links depth-aware synthetic fog generation, image restoration, object detection, and tracking within a unified pipeline. Given the practical difficulty of collecting and annotating foggy UAV scenes, synthetic fog is generated from real clear-weather outdoor images containing UAV targets using monocular depth estimation and the atmospheric scattering model. Representative restoration methods from classical, convolutional neural network (CNN)-based, and transformer-based families are first compared, after which the selected restoration model is integrated into the downstream perception pipeline. Detection is evaluated under both clean-only and fog-inclusive training regimes using multiple detector variants, while tracking-by-detection is assessed on clean, foggy, and restored video sequences. Beyond image-level restoration metrics, the study evaluates how fog and restoration affect detection robustness and tracking performance. The results show that fog substantially degrades both detection and tracking, primarily through increased missed detections. Fog-inclusive training provides the most consistent improvement in robustness, whereas test-time restoration is most beneficial when the detector has been trained only on clean imagery. These findings show that restoration quality does not necessarily translate into proportional gains in downstream perception and therefore should be evaluated jointly with detection and tracking performance.
Adwait Chandorkar, Kai Krink, Yerdana Maulenbay +2cs.CV
Recent advancements in LiDAR-only 3D object detection have demonstrated improved detection accuracy over benchmark datasets. However, the adversarial robustness of these models remains untested. Very few adversarial robustness studies exist for LiDAR-only 3D object detection and unfortunately, even they are limited to legacy models. Moreover, there is a systemic gap in the existing evaluation frameworks that rely simply on mAP ignoring other structural and predictive factors. To fill this gap, we propose a holistic framework that evaluates adversarial robustness using two structural factors (point cloud density and point cloud localization) and three predictive factors (misclassification, localization error, distance from ego). Using this framework, we perform an empirical study and critical analysis on recent and legacy state-of-the-art models using adversarial attacks specifically designed for LiDAR-based models. Our key finding is that high-capacity, voxel-based detectors are more susceptible to structured coordinate perturbations than pillar-based detectors. Additionally, non-anchor-based detectors demonstrate poor adversarial robustness, which necessitates rethinking model training techniques. Overall, our results demonstrate that recent models are as vulnerable to adversarial attacks as their predecessors. Therefore, we argue that there is a need to improve the evaluation benchmarks for 3D object detection that not only reward architectural modifications for improving detection accuracy, but also evaluate whether the design choices improve adversarial robustness.
Jonas Klotz, Cassio F. Dantas, Pallavi Jain +2cs.CV cs.AI
Sparse autoencoders (SAEs) are increasingly used to extract interpretable concepts from vision and vision language models, yet existing evaluation methods largely rely on proxy metrics or qualitative inspection rather than measuring semantic correspondence. We present a human-grounded evaluation framework that quantifies alignment between SAE latents and human-annotated concepts, without requiring user studies, and validate this matching through targeted attribute perturbations. To enable this intervention-style evaluation in vision, we construct synCUB and synCOCO, synthetic benchmarks of paired images that differ in exactly one attribute. We introduce Fully-Binary Matching Pursuit (FBMP), a coalition-based matching procedure that supports many-to-one mappings between SAE latents and annotated concepts, and consistently outperforms one-to-one baselines. For functional validation, we propose a Targeted Attribute Perturbation Alignment Score (TAPAScore), which tests whether matched concepts respond selectively and in the expected direction under targeted image-level attribute perturbations. Under sanity checks, our matching and TAPAScore are the only evaluated metrics that reliably distinguish trained SAEs from untrained ones. Across SAEs trained on CLIP and DINOv2 embeddings, we find that increased overcompleteness can reduce perturbation alignment, indicating a reduction in interpretability. Our evaluation framework suggests that moderate dictionary sizes provide the best trade-off, yielding the most interpretable SAEs. Code and datasets are available at https://github.com/JonasKlotz/sae-concept-eval.
As video generation models like Veo 3.1 and LTX-2 advance, their ability to accurately represent diverse global cultures remains a critical yet understudied frontier. Current metrics, such as VideoScore, only measure visual quality but offer no mechanism for assessing cultural faithfulness. Consequently, a model that replaces a Namaste with a handshake receives the same score as one that generates the gesture correctly. We propose CultureScore, a compositional evaluation framework that decomposes cultural faithfulness into three granular dimensions: Identity (who is represented), Context (culturally localized background), and Behavior (normative gestures and interactions). We operationalize this framework through an evaluation suite spanning 10 countries, yielding 6,174 generated videos across three state-of-the-art models. Our evaluation reveals that no current model achieves culturally faithful video generation: the best-performing model reaches only 56.8\% overall CultureScore, with Behavior the most challenging dimension, which remains below 52.1\% across all models. Furthermore, human preference rankings align directionally with CultureScore but are inverted relative to VideoScore; the highest-scoring model on visual quality was ranked last by annotators, underscoring that cultural faithfulness is an essential criterion for equitable video generation. Data and code are available at https://huggingface.co/datasets/ankurani/CultureScore.
In most real-world image-to-image (I2I) scenarios, existing evaluations primarily focus on instruction following and the perceptual quality or aesthetics of the generated images. However, they largely fail to assess whether the output image preserves the semantic correspondence and spatial structure of the input image. To address this limitation, we propose StableI2I, a unified and dynamic evaluation framework that explicitly measures content fidelity and pre--post consistency across a wide range of I2I tasks without requiring reference images, including image editing and image restoration. In addition, we construct StableI2I-Bench, a benchmark designed to systematically evaluate the accuracy of MLLMs on such fidelity and consistency assessment tasks. Extensive experimental results demonstrate that StableI2I provides accurate, fine-grained, and interpretable evaluations of content fidelity and consistency, with strong correlations to human subjective judgments. Our framework serves as a practical and reliable evaluation tool for diagnosing content consistency and benchmarking model performance in real-world I2I systems.
Video generation models have developed rapidly in recent years, where generating natural human motion plays a pivotal role. However, accurately evaluating the quality of generated human motion video remains a significant challenge. Existing evaluation metrics primarily focus on global scene statistics, often overlooking fine-grained human details and consequently failing to align with human subjective preference. To bridge this gap, we propose HuM-Eval, a novel human-centric evaluation framework that adopts a coarse-to-fine strategy. Specifically, our framework first utilizes a Vision Language Model to perform a coarse assessment of global video quality. It then proceeds to a fine-grained analysis, using 2D pose to verify anatomical correctness and 3D human motion to evaluate motion stability. Extensive experiments demonstrate that HuM-Eval achieves an average human correlation of 58.2%, outperforming state-of-the-art baselines. Furthermore, we introduce HuM-Bench, a comprehensive benchmark comprising 1,000 diverse prompts, and conduct a detailed evaluation of existing text-to-video models, paving the way for next-generation human motion generation.