Pith. sign in

REVIEW 4 major objections 6 minor 36 references

Advancing Vision-based Human Action Recognition: Exploring Vision-Language CLIP Model for Generalisation in Domain-Independent Tasks

T0 review · 4 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper evaluates CLIP's zero-shot action recognition under masking and claims that a per-class learned noise vector added to its frozen features reduces label dispersion, raising accuracy and confidence.

desk verdict Exploratory CLIP masking study with an unvalidated and internally flawed class-specific noise method; the descriptive histogram analysis has some value, but the central claim does not hold. read the letter →

arxiv 2507.18675 v2 pith:HIT6RGRK submitted 2025-07-24 cs.CV cs.LG

Utkarsh Shandilya , Marsha Mariya Kappan , Sanyam Jain , Vijeta Sharma This is my paper · ORCID
classification cs.CVcs.LG
keywords CLIPhumanactionrecognitionUCF-101zero-shotclassificationlabeldispersionclass-specificnoisetripletlossmaskingstrategies
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper evaluates the CLIP vision-language model on the UCF-101 human action recognition dataset under three kinds of perturbation: random black masking at 10, 30, and 50 percent, feature-specific masking that removes background elements, and isolation masking that keeps only class-defining regions. The reported histograms show that CLIP frequently misclassifies actions and spreads its predictions across many labels, a behaviour the paper calls label dispersion. To counter this, the authors propose learning a class-specific noise vector for each action class, optimized with a triplet loss and added to CLIP's feature representation, so that the model attends to class-defining features. They claim this augmentation improves classification accuracy and model confidence while reducing bias. The stated motivation is healthcare applications such as fall detection, patient monitoring, and surgical skill assessment, where action recognition must generalize across domains.

What carries the argument

The load-bearing mechanism is an additive class-specific noise vector $N_c$, one per action class, stored in a dictionary $D_c$ and added to CLIP's extracted feature representation: $F_c = f(X_c) + N_c$. The noise is learned with a triplet loss that pulls augmented anchor and positive frame features from the same class together while pushing apart features from a different class, with a margin $\alpha$. In the masking experiments, the Segment Anything Model (SAM) supplies the masks used for feature-specific and isolation masking. The noise is doing the work: it is the only proposed remedy for the label dispersion the masking tasks expose.

What would settle it

Learn $N_c$ on a designated subset of UCF-101 frames per class using the described triplet loss, then evaluate on a disjoint held-out subset of the same classes, with and without 10/30/50 percent black masking, and compare top-1 accuracy and average confidence against plain CLIP. If the noise-augmented version does not beat plain CLIP on held-out frames, the claimed mechanism fails.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central discovery is that CLIP's zero-shot action recognition on UCF-101 is unstable under masking: random black masking at 10–50 percent scatters predictions across unrelated classes, feature-specific masking of backgrounds such as grass or pitch leads to complete misclassification of Cricket Shot, and isolation masking does not recover the ground truth. The paper then proposes class-specific noise $N_c$, learned per class through a triplet loss and added to the frozen visual features as $F_c = f(X_c) + N_c$, and asserts in Section 5.4 that this augmentation helps the model prevent label dispersion and produce high-confidence predictions. The claim is that adding this per-class learned perturbation to CLIP's features reinforces class-defining cues and improves both label accuracy and confidence, on ordinary frames and on SAM-segmented frames alike.

Load-bearing premise

The load-bearing premise is that a noise vector $N_c$, learned from a small set of frames of class $c$ with a triplet loss, transfers to other frames of the same class and can be added to CLIP's frozen features to improve zero-shot predictions on those other frames.

Editorial extensions

If this is right

  • On UCF-101, plain CLIP's predictions scatter across many labels and misclassify actions when key cues are blacked out or backgrounds are removed; the paper treats this as label dispersion.
  • If class-specific noise works as claimed, CLIP can be strengthened for action recognition without retraining the vision encoder, since the noise is added to frozen features.
  • The proposed pipeline is aimed at healthcare scenarios such as fall detection, patient monitoring, and surgical skill assessment, where partial occlusion and domain shift are common.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper's own evidence for the noise augmentation is a single illustrative figure with no training details or quantitative comparison; the claim should be read as proposed and illustrated rather than established.
  • Because the noise $N_c$ is an additive vector in feature space, it is equivalent to a per-class prototype shift; that invites a direct comparison against prototype-based few-shot action recognition methods.
  • The paper's 'domain-independent generalization' is not tested, since all experiments use UCF-101; an obvious test is to learn noise dictionaries on one dataset and evaluate on HMDB51 or Kinetics-400.
  • A concrete robustness test would apply the same learned noise dictionary under unseen mask patterns or object occlusions to see whether the improvement persists beyond the perturbations studied.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper evaluates the CLIP vision-language model on the UCF-101 human action recognition dataset under three masking strategies (random percentage/shape-based black masking, feature-specific masking via SAM, and isolation masking that retains only class-specific regions), and proposes a class-specific noise (CSN) augmentation learned via a triplet loss to improve classification accuracy, confidence, and label dispersion. The results are presented primarily as frequency histogram charts and qualitative observations, with full prediction tables for Task 1 in the appendix. The paper concludes by discussing challenges and future directions for applying CLIP to healthcare-related action recognition.

Significance. If the proposed class-specific noise augmentation were rigorously validated, it could offer a simple plug-in mechanism to improve zero-shot CLIP robustness under occlusion and domain shift, which would be relevant to healthcare monitoring applications. The paper also contributes a useful descriptive evaluation of CLIP on UCF-101 under various masking perturbations, and it makes its code and full per-class prediction tables publicly available. However, the central claim of improved accuracy, confidence, and reduced bias is supported only by a few qualitative histograms, with no quantitative metrics, no comparison baselines, no error bars, and no clearly defined inference protocol. The method as described in Section 4 is also internally degenerate, as the triplet loss does not actually depend on the noise for the anchor-positive pair. As a result, the paper does not, in its current form, provide a reliable basis for its headline conclusions.

major comments (4)
  1. [§4 and §5.4, Class-Specific Noise] The inference protocol for the proposed class-specific noise is undefined. The noise N_c is learned per class and stored in dictionary D_c, but at test time the action class is unknown. The paper never states how N_c is selected for a query frame. If the ground-truth label is used to retrieve N_c (as the 'GT Action Image' panel in Figure 6 suggests), the improvement is label leakage; if the model's own top-1 prediction is used to select N_c, the augmentation is circular and merely reinforces the prior guess. This makes the claimed gains in accuracy and confidence in Section 5.4 impossible to interpret or reproduce.
  2. [§4, Triplet Loss Equation] The triplet loss is degenerate as written. Both the anchor and the positive sample are augmented with the same noise N_c, so d(f(X_c)+N_c, f(X_positive)+N_c) = d(f(X_c), f(X_positive)), which is independent of N_c. Only the negative distance depends on N_c, so minimizing the loss can arbitrarily increase the norm of N_c or push it toward the negative class without enforcing intra-class compactness. The paper's claim that this loss 'ensures that the noise N_c is optimized to highlight class-specific features' is therefore not supported by the stated objective.
  3. [§5.4 and Figure 6] The central claim that class-specific noise improves accuracy, confidence, and reduces bias is supported only by qualitative top-3 histogram examples. No accuracy numbers, no baseline CLIP comparison, no error bars, and no statistical analysis are provided. The paper itself states in §5.4 that 'more experiments can be carried using tasks 1 to 4 in order to verify methods for task 5,' which explicitly acknowledges that the method has not been properly evaluated. Without quantitative results, the abstract's assertion of 'improved classification accuracy and model confidence' is unsubstantiated.
  4. [§5.3 (Tasks 3 and 4)] The feature-specific and isolation masking experiments are conducted on a single action class (Cricket Shot), and the paper notes that 'none of the segmented images were labelled as the ground truth.' This single-class, all-failure result is presented without any quantitative summary and without extending to other classes, so it cannot support the paper's broader generalization claims about CLIP's behavior under masking.
minor comments (6)
  1. [§2] There are several incomplete or informal sentences, e.g., 'In s study by [7], evaluated the effectiveness of CLIP' and '[2]' in the references has no venue. The prose would benefit from a careful editing pass.
  2. [§5.2] The abstract and Task 2 description mention 'shape-based' black masking, but the implementation describes only random percentage-based pixel masking. The paper should clarify whether shape-based masking was actually performed and, if so, how the shapes were generated.
  3. [Table 4 (Appendix)] The caption states 'with this particular experiment we kept only four labels in the ground truth labels for CLIP to take as query.' This is a substantial restriction on the label space that is not mentioned in the main text, and it makes the Task 2 histogram results non-comparable to standard 101-way classification. This should be stated clearly in the main text.
  4. [§5.1 and Tables 1-3] The tables in the appendix are useful, but the paper does not provide a summary statistic such as top-1 accuracy over the 101 classes. Adding mean accuracy and confidence would make these tables more interpretable.
  5. [Figure 6] The figure is difficult to read because the panels are small and the histograms are not clearly labeled. A higher-resolution figure with explicit axis labels and a clearer description of the three prediction conditions would improve reproducibility.
  6. [§4, Notation] The notation for the augmented features is inconsistent: F_c, F_aug, and F_fine-tuned are used for essentially the same operation f(X) + N_c. Please use a single consistent symbol and define all variables explicitly.

Circularity Check

2 steps flagged · score 6.0 of 10

The claimed class-specific-noise improvement is not an independent prediction: N_c is fit per ground-truth class with no test-time selection rule, so the gain is either label leakage or self-reinforcement; the triplet-loss positive term also cancels N_c.

  1. fitted input called prediction [Section 4, Task 5 ('Class-Specific Noise'); Section 5.4, Figure 6]
    "For each class c, a dictionary Dc is constructed, storing the characteristic features of that class. These features are augmented by learnable noise Nc for each class... The pretrained models can access the class-specific noise through the dictionary Dc either as an augmentation to the dataset or as a learnable parameter. Fig. 6. Results show that class specific noise enhanced the predictions for both the cases."

    The noise N_c is fit per ground-truth class from that class's own frames, but no rule is given for selecting N_c for an unlabeled test frame. The only result shown (Figure 6, 'Ground Truth Action Image' versus 'CLIP + CSN Augment') applies the per-class noise to frames whose class is already known. If the ground-truth label is used to retrieve N_c, the reported top-3 improvement is label leakage: the class is inserted before inference, so the 'prediction' is forced by construction. If instead the model's own top-1 prediction selects N_c, the added vector merely reinforces the prior guess and the histogram is circular. Either way, the claimed accuracy gain is not an independent prediction.

  2. other [Section 4, 'Triplet Loss Integration' (Laug equation)]
    "The augmented triplet loss incorporating the class-specific noise can be written as: Laug = max(0, d(f(X_c)+N_c, f(X_positive)+N_c) - d(f(X_c)+N_c, f(X_negative)+N_c') + alpha)."

    In the stated equation, the anchor and positive both add the same N_c, so the positive distance reduces to d(f(X_c), f(X_positive)) and N_c cancels. The loss therefore cannot minimize intra-class distance via the noise; N_c affects only the negative distance and can grow unboundedly to satisfy the margin. The claimed mechanism by which class-specific noise highlights class-defining features is absent from the stated objective, so the Figure 6 improvements cannot be attributed to the proposed loss. This is a reduction of the method's own equation, though it is a degeneracy rather than a classical input-output circularity.

full rationale

Tasks 1-4 are descriptive evaluations of CLIP on UCF-101 frames under masking and segmentation; they are not circular because they report CLIP's external behavior against the frozen model and dataset. The circularity is confined to Task 5: the class-specific noise is a per-class parameter fit with ground-truth class labels, and the claimed improvement is displayed on the same known classes without any test-time selection rule, making the central 'prediction' either label leakage (ground-truth retrieval) or self-reinforcement (predicted-label retrieval). The triplet-loss positive term also cancels N_c, so the stated learning signal does not implement the claimed intra-class compactness. No load-bearing self-citation or imported uniqueness theorem appears; reference [13] (a co-author) is only a future-work pointer. Because the paper gives no quantitative comparison and no inference protocol, the defect is partial circularity rather than a fully forced derivation; hence score 6.

Assumptions & free parameters 1 free parameters · 3 assumptions · 1 invented entities

The central claim rests on the assumption that a learnable noise vector per class can be trained with triplet loss on a few frames and then improve zero-shot CLIP predictions. The paper provides no independent evidence for this, and the noise itself is an invented construct without external validation.

free parameters (1)
  • triplet loss margin alpha
    The margin for triplet loss is a hyperparameter; its value is not specified.
assumptions (3)
  • domain assumption CLIP's pretrained embeddings are suitable for zero-shot human action recognition from static frames.
    The entire evaluation assumes that CLIP's image encoder can recognize actions from individual frames without temporal information (Section 3.3).
  • domain assumption SAM segmentation accurately separates class-specific from non-class-specific features.
    Tasks 3 and 4 rely on Segment Anything Model masks to remove background, with hand-picked coordinates (Section 5.3).
  • domain assumption Triplet loss distances in CLIP embedding space reflect action class similarity.
    The class-specific noise is trained to minimize or maximize Euclidean distances in feature space (Section 4).
invented entities (1)
  • Class-specific noise N_c
    purpose: A learnable per-class offset added to CLIP image features to improve class separability and robustness to masking.
    The paper introduces this construct but provides no falsifiable prediction or external validation; it is not shown to transfer to new data.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Advancing Vision-based Human Action Recognition: Exploring Vision-Language CLIP Model for Generalisation in Domain-Independent Tasks." pith.science (2026). https://pith.science/paper/HIT6RGRK

@misc{pith2026250718675,
  author       = {Pith},
  title        = {Pith review of: Advancing Vision-based Human Action Recognition: Exploring Vision-Language CLIP Model for Generalisation in Domain-Independent Tasks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HIT6RGRK}},
  note         = {Machine review of arXiv:2507.18675}
}
read the original abstract

Human action recognition plays a critical role in healthcare and medicine, supporting applications such as patient behavior monitoring, fall detection, surgical robot supervision, and procedural skill assessment. While traditional models like CNNs and RNNs have achieved moderate success, they often struggle to generalize across diverse and complex actions. Recent advancements in vision-language models, especially the transformer-based CLIP model, offer promising capabilities for generalizing action recognition from video data. In this work, we evaluate CLIP on the UCF-101 dataset and systematically analyze its performance under three masking strategies: (1) percentage-based and shape-based black masking at 10%, 30%, and 50%, (2) feature-specific masking to suppress bias-inducing elements, and (3) isolation masking that retains only class-specific regions. Our results reveal that CLIP exhibits inconsistent behavior and frequent misclassifications, particularly when essential visual cues are obscured. To overcome these limitations, we propose incorporating class-specific noise, learned via a custom loss function, to reinforce attention to class-defining features. This enhancement improves classification accuracy and model confidence while reducing bias. We conclude with a discussion on the challenges of applying such models in clinical domains and outline directions for future work to improve generalizability across domain-independent healthcare scenarios.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 25 canonical work pages

  1. [1]

    Leveraging Vision -Language Models for Improving Domain Generalization in Image Classification

    Sravanti Addepalli et al. “Leveraging Vision -Language Models for Improving Domain Generalization in Image Classification”. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024, pp. 23922–23932

    work page 2024

  2. [2]

    Are Visual-Language Models Effective Action Recognition? A Comparative Study

    Mahmoud Ali, Di Yang, and François Brémond. “Are Visual-Language Models Effective Action Recognition? A Comparative Study”. In: (2023)

    work page 2023

  3. [3]

    Vision–language model for visual question answering in medical imagery

    Yakoub Bazi et al. “Vision–language model for visual question answering in medical imagery”. In: Bioengineering 10.3 (2023), p. 380

    work page 2023

  4. [4]

    When deep learners change their mind: Learning dynamics for active learning

    Javad Zolfaghari Bengar, Bogdan Raducanu, and Joost van de Wei- jer. “When deep learners change their mind: Learning dynamics for active learning”. In: International Conference on Computer Analysis of Images and Patterns. Springer. 2021, pp. 403–413

    work page 2021

  5. [5]

    An introduction to vision -language model- ing

    Florian Bordes et al. “An introduction to vision -language model- ing”. In: arXiv preprint arXiv:2405.17247 (2024)

    arXiv 2024

  6. [6]

    PracticalDG: Perturbation Distillation on Vision- Language Models for Hybrid Domain Generalization

    Zining Chen et al. “PracticalDG: Perturbation Distillation on Vision- Language Models for Hybrid Domain Generalization”. In: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024, pp. 23501–23511

    work page 2024

  7. [7]

    Pub - medclip: How much does clip benefit visual question answering in the medical domain?

    Sedigheh Eslami, Christoph Meinel, and Gerard De Melo. “Pub - medclip: How much does clip benefit visual question answering in the medical domain?” In: Findings of the Association for Compu- tational Linguistics: EACL 2023. 2023, pp. 1181–1193

    work page 2023

  8. [8]

    Clipsyntel: clip and llm synergy for multi - modal question summarization in healthcare

    Akash Ghosh et al. “Clipsyntel: clip and llm synergy for multi - modal question summarization in healthcare”. In: Proceedings of the AAAI Conference on Artificial Intelligence. Vol. 38. 20. 2024, pp. 22031–22039

    work page 2024

Show all 36 references
  1. [9]

    Learn2augment: learning to composite videos for data augmentation in action recognition

    Shreyank N Gowda et al. “Learn2augment: learning to composite videos for data augmentation in action recognition”. In: European conference on computer vision. Springer. 2022, pp. 242–259

    2022

  2. [10]

    Class-Specific Noise Injection for Improved Road Segmentation

    Yukai Gu et al. “Class-Specific Noise Injection for Improved Road Segmentation”. In: International Conference on Machine Learn - ing and Intelligent Communications. Springer. 2023, pp. 96–106

    2023

  3. [11]

    Temporal Modeling Approach for Video Action Recognition Based on Vision-language Models

    Yue Huang and Xiaodong Gu. “Temporal Modeling Approach for Video Action Recognition Based on Vision-language Models”. In: International Conference on Neural Information Processing. Springer. 2023, pp. 512–523

    2023

  4. [12]

    Perturbation- based methods for explaining deep neural networks: A survey

    Maksims Ivanovs, Roberts Kadikis, and Kaspars Ozols. “Perturbation- based methods for explaining deep neural networks: A survey”. In: Pattern Recognition Letters 150 (2021), pp. 228–234

    2021

  5. [13]

    A dversarial attack on yolov5 for traffic and road sign detection

    Sanyam Jain. “A dversarial attack on yolov5 for traffic and road sign detection”. In: 2024 4th International Conference on Applied Artificial Intelligence (ICAPAI). IEEE. 2024, pp. 1–5. Title Suppressed Due to Excessive Length 13

    2024

  6. [14]

    Scaling up visual and vision-language representa- tion learning with noisy text supervision

    Chao Jia et al. “Scaling up visual and vision-language representa- tion learning with noisy text supervision”. In: International con- ference on machine learning. PMLR. 2021, pp. 4904–4916

    2021

  7. [15]

    Language augmentation in clip for improved anatomy detection on multi-modal medical images

    Mansi Kakkar et al. “Language augmentation in clip for improved anatomy detection on multi-modal medical images”. In: 2024 46th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC). IEEE. 2024, pp. 1–4

    2024

  8. [16]

    Lalm: Long-term action anticipation with language models

    Sanghwan Kim et al. “Lalm: Long-term action anticipation with language models”. In: arXiv preprint arXiv:2311.17944 (2023)

    2023 arXiv

  9. [17]

    Segment Anything

    Alexander Kirillov et al. Segment Anything. 2023. arXiv: 2304. 02643 [cs.CV]. URL: https://arxiv.org/abs/2304.02643

    2023 arXiv

  10. [18]

    Qilin-med-vl: Towards chinese large vision- language model for general healthcare

    Junling Liu et al. “Qilin-med-vl: Towards chinese large vision- language model for general healthcare”. In: arXiv preprint arXiv:2310.17956 (2023)

    2023 arXiv

  11. [19]

    Enhancing clip with gpt-4: Harness- ing visual descriptions as prompts

    Mayug Maniparambil et al. “Enhancing clip with gpt-4: Harness- ing visual descriptions as prompts”. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. 2023, pp. 262–271

    2023

  12. [20]

    Learning transferable visual models from nat- ural language supervision

    Alec Radford et al. “Learning transferable visual models from nat- ural language supervision”. In: International conference on ma - chine learning. PMLR. 2021, pp. 8748–8763

    2021

  13. [21]

    Vision language models are blind

    Pooyan Rahmanzadehgervi et al. “Vision language models are blind”. In: arXiv preprint arXiv:2407.06581 (2024)

    2024 arXiv

  14. [22]

    How much can clip benefit vision-and-language tasks?

    Sheng Shen et al. “How much can clip benefit vision-and-language tasks?” In: arXiv preprint arXiv:2107.06383 (2021)

    2021 arXiv

  15. [23]

    Test-time prompt tuning for zero-shot general- ization in vision-language models

    Manli Shu et al. “Test-time prompt tuning for zero-shot general- ization in vision-language models”. In: Advances in Neural Infor- mation Processing Systems 35 (2022), pp. 14274–14289

    2022

  16. [24]

    Safeguarding Vision-Language Models Against Patched Visual Prompt Injectors

    Jiachen Sun et al. “Safeguarding Vision-Language Models Against Patched Visual Prompt Injectors”. In: arXiv preprint arXiv:2405.10529 (2024)

    2024 arXiv

  17. [25]

    Motionclip: Exposing human motion genera - tion to clip space

    Guy Tevet et al. “Motionclip: Exposing human motion genera - tion to clip space”. In: European Conference on Computer Vision. Springer. 2022, pp. 358–374

    2022

  18. [26]

    Xraygpt : Chest radiographs summariza - tion using medical vision -language models

    Omkar Thawkar et al. “Xraygpt : Chest radiographs summariza - tion using medical vision -language models”. In: arXiv preprint arXiv:2306.07971 (2023)

    2023 arXiv

  19. [27]

    CLIP with Quality Captions: A Strong Pretraining for Vision Tasks

    Pavan Kumar Anasosalu Vasu et al. “CLIP with Quality Captions: A Strong Pretraining for Vision Tasks”. In: arXiv preprint arXiv:2405.08911 (2024)

    2024 arXiv

  20. [28]

    Actionclip: A new paradigm for video action recognition

    Mengmeng Wang, Jiazheng Xing, and Yong Liu. “Actionclip: A new paradigm for video action recognition”. In: arXiv preprint arXiv:2109.08472 (2021)

    2021 arXiv

  21. [29]

    Actionclip: Adapting language-image pre- trained models for video action recognition

    Mengmeng Wang et al. “Actionclip: Adapting language-image pre- trained models for video action recognition”. In: IEEE Transac- tions on Neural Networks and Learning Systems (2023). 14 U. Shandilya et al

    2023

  22. [30]

    Incorporating Scene Graphs into Pre-trained Vision-Language Models for Multimodal Open-vocabulary Action Recognition

    Chao Wei and Zhidong Deng. “Incorporating Scene Graphs into Pre-trained Vision-Language Models for Multimodal Open-vocabulary Action Recognition”. In: 2024 IEEE International Conference on Robotics and Automation (ICRA). IEEE. 2024, pp. 440–447

    2024

  23. [31]

    Clipself: Vision transformer distills itself for open- vocabulary dense prediction

    Size Wu et al. “Clipself: Vision transformer distills itself for open- vocabulary dense prediction”. In: arXiv preprint arXiv:2310.01403 (2023)

    2023 arXiv

  24. [32]

    Revisiting classi - fier: Transferring vision-language models for video recognition

    Wenhao Wu, Zhun Sun, and Wanli Ouyang. “Revisiting classi - fier: Transferring vision-language models for video recognition”. In: Proceedings of the AAAI conference on artificial intelligence. Vol. 37. 3. 2023, pp. 2847–2855

    2023

  25. [33]

    Investigating Compositional Challenges in Vision- Language Models for Visual Grounding

    Yunan Zeng et al. “Investigating Compositional Challenges in Vision- Language Models for Visual Grounding”. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion. 2024, pp. 14141–14151

    2024

  26. [34]

    PeVL : Pose -Enhanced Vision-Language Model for Fine-Grained Human Action Recognition

    Haosong Zhang et al. “PeVL : Pose -Enhanced Vision-Language Model for Fine-Grained Human Action Recognition”. In: Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024, pp. 18857–18867

    2024

  27. [35]

    Vision -language models for vision tasks: A survey

    Jingyi Zhang et al. “Vision -language models for vision tasks: A survey”. In: IEEE Transactions on Pattern Analysis and Machine Intelligence (2024)

    2024

  28. [36]

    Clip in medical imaging: A comprehensive sur- vey

    Zihao Zhao et al. “Clip in medical imaging: A comprehensive sur- vey”. In: arXiv preprint arXiv:2312.07353 (2023). Appendix In this section we outline all class wise results for task 1 in Tables 1, 2 and 3. Moreover we also release handpicked interesting results in Table 4. Ti...

    2023 arXiv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.