Pith. sign in

REVIEW 4 major objections 7 minor 66 references

Task-Adapter++: Task-specific Adaptation with Order-aware Alignment for Few-shot Action Recognition

T0 review · 4 major / 7 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 Task-Adapter++ shows few-shot action recognition improves by freezing CLIP and tuning tiny order-aware adapters.

desk verdict Solid FSAR paper where the visual task-adaptation idea is the real contribution; the semantic stage-alignment story needs a diagnostic before the SOTA claim fully lands. read the letter →

arxiv 2505.06002 v2 pith:7VGI7AL7 submitted 2025-05-09 cs.CV

Congqi Cao , Peiheng Han , Yueran zhang , Yating Yu , Qinyi Lv , Lingtong Min , Yanning zhang This is my paper · ORCID
classification cs.CV
keywords few-shotactionrecognitionparameter-efficientfine-tuningCLIPadaptationtask-specificsemanticordercross-modalalignmentLLM-generateddescriptions
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

The paper claims that few-shot action recognition can be improved by keeping a pre-trained CLIP model frozen and inserting small trainable adapters into both its visual and text branches, instead of full fine-tuning. It introduces a Task-Adapter inside the visual encoder that lets support and query videos attend to each other at shared spatio-temporal locations, highlighting the most task-discriminative cues. On the text side, it uses LLM-generated ordered sub-action descriptions (beginning, process, end) and a semantic order adapter to model the sequence. A fine-grained cross-modal alignment then segments aligned query features into temporal stages and matches them stage-wise with the sub-action semantics. The combination is reported to reach new state-of-the-art results on five benchmarks while training only 3.5M or 7.5M parameters.

What carries the argument

The central objects are the task adapter and the semantic order adapter. The task adapter reuses the frozen MSA layer as a Task-MSA that performs self-attention across the same spatio-temporal positions across all videos in an episode, so the most discriminative task-specific information is emphasized during feature extraction. The semantic order adapter reuses the frozen MSA as an O-MSA to model the sequential relationships among the three ordered sub-action description prompts, injecting order information into the text features. The fine-grained cross-modal alignment applies cross-attention between adjacent frames, producing aligned features of length $T-1$, then splits them into three overlapping segments--frames 1-3, 3-5, and 5-7--and matches each segment by cosine similarity with the corresponding LLM-generated sub-action feature, combining the two matching scores by multiplication in Eq. (18).

What would settle it

Take a few-shot episode and permute the order of the three LLM sub-action descriptions--for example, swap the end description with the beginning description--before computing the cross-modal term in Eq. (18); if accuracy stays nearly the same, the order-aware stage correspondence is not doing the claimed work, while a large drop would confirm the alignment matters. A second check is to replace the fixed equal thirds of Eq. (16) with manually labeled temporal phase boundaries and see whether accuracy rises.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes that parameter-efficient dual adaptation--task-specific adapters in the visual branch and order-aware adapters in the text branch of a frozen CLIP--consistently outperforms both full fine-tuning and prior PEFT methods on SSv2-Small, SSv2-Full, HMDB51, UCF101, and Kinetics. The principal novelty claimed is performing task-specific adaptation during spatio-temporal feature extraction: a reused frozen MSA layer acts as a Task-MSA that runs cross-video attention over tokens at the same spatio-temporal location, with learnable adapters around it. The semantic branch uses a semantic order adapter with a reused O-MSA to model the relationships among three LLM-generated sub-action descriptions. Finally, a fine-grained cross-modal alignment module applies one cross-attention step between adjacent frames, divides the aligned features into three temporal segments, and matches each segment to the corresponding sub-action semantic feature; the final classification score is the element-wise product of the video matching score and the video-text matching score.

Load-bearing premise

Cross-modal matching assumes that the LLM's three ordered sub-action descriptions correspond, in order, to thirds of the query video, yet the video is divided into three equal overlapping segments without evidence that the LLM's beginning, process, and end stages line up with those frame intervals.

Editorial extensions

If this is right

  • If the central claim holds, frozen-CLIP few-shot action recognition can reach higher accuracy than full fine-tuning while reducing trainable parameters from about 149.6M to 3.5M or 7.5M.
  • The method is metric-agnostic: it is reported to improve performance when combined with TRX, OTAM, Bi-MHM, or simple prototype averaging, meaning the visual adapter carries most of the gain.
  • Decomposing an action label into ordered sub-action descriptions and matching stage-wise outperforms averaging the descriptions or concatenating them into one prompt, supporting the paper's claim that semantic order matters.
  • On temporal-heavy datasets like SSv2, inserting the task adapter into the top six layers works best; on scene-heavy datasets like Kinetics, UCF101, and HMDB51, the top two layers suffice, yielding different parameter counts.
  • The video-text matching branch alone improves from 50.4% to 54.3% after adding order adaptation and fine-grained alignment, indicating the semantic branch becomes informative rather than redundant.

Reading between the lines

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

  • The fixed equal-thirds segmentation of the 8-frame video is a crude stand-in for actual action phases; if the LLM's beginning/process/end descriptions were aligned via learned temporal boundaries or annotated phase labels, the cross-modal term might become more reliable and accurate.
  • The dual-adapter recipe likely transfers to other video-language problems such as temporal action localization, video captioning, or step-level instruction understanding, wherever ordered text descriptions exist alongside video.
  • Because only adapters are trained, the approach could be paired with other frozen backbones or extended toward zero-shot regimes, though the paper does not test those settings.
  • The product rule for combining video and video-text scores may suppress correct predictions when one branch is noisy; a learned fusion or sum rule could be more robust, and the paper does not compare these alternatives.
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 / 7 minor

Summary. The paper proposes Task-Adapter++, a parameter-efficient framework for few-shot action recognition built on frozen CLIP. It inserts task adapters with a reused task-specific MSA into the last L layers of the visual encoder, uses GPT-4 to generate three ordered sub-action descriptions per class and models them with order adapters in the text encoder, and computes a fine-grained cross-modal alignment score by splitting aligned query features into three temporal segments and matching them to the three semantic features. The final classification score is the product of the video-prototype matching score and the cross-modal matching score. Experiments on SSv2-Small, SSv2-Full, HMDB51, UCF101, and Kinetics report state-of-the-art results with 3.5M or 7.5M trainable parameters.

Significance. Assuming the semantic-stage alignment is validated, the paper makes a solid contribution: it demonstrates that a frozen CLIP backbone with lightweight task adapters and LLM-generated sub-action descriptions can outperform full fine-tuning baselines on five few-shot action recognition benchmarks using only 3.5M or 7.5M trainable parameters. The component-wise ablations in Tables 3 and 4 are informative, the comparison across metric modules in Fig. 10 is useful, and the open-sourced code supports reproducibility. The main risk is that the cross-modal gain rests on the unverified correspondence between LLM descriptions and fixed temporal segments, and the current text does not fully explain or validate that correspondence.

major comments (4)
  1. [3.6, Eqs. (15)-(17), Table 4] The stage-wise cross-modal score rests on an unverified correspondence between the three GPT-4 sub-action descriptions and the three temporal segments defined in Eq. (16). First, the indexing of the aligned features is ambiguous: Eq. (15) defines aligned features from frames i-1 and i, so with T=8 the aligned features are naturally indexed as i=2,...,8, whereas Eq. (16) sums i=1..3, 3..5, and 5..7; if instead the aligned feature index is meant to denote the pair (i,i+1), the segments overlap at pairs 3 and 5, so the three segments cover frames 1-4, 3-6, and 5-8. Neither the overlap nor the boundary choice is justified. More importantly, the paper provides no evidence that the LLM-generated 'beginning, process, end' descriptions correspond to these fixed intervals across classes with highly variable action durations; Table 4 shows that adding cross-modal alignment without the sub-action descriptions decreases accuracy from 60.2 to 58.5, so the gain of the full model is contingent on the semantic-stage correspondence being correct. Please provide a corrected and unambiguous definition of the segments, an ablation with non-overlapping or learned/adaptive segmentation, and a diagnostic or control such as shuffling the three sub-action descriptions or annotating temporal boundaries for a sample of classes.
  2. [Table 4] The third row of Table 4 is difficult to interpret because the text does not specify what the text branch contains when the 'Three Sub-actions' component is off. Equation (17) requires three semantic features to match the three visual segments, but Section 3.5 only defines those features from the three sub-action prompts; if cross-modal alignment is applied without sub-action descriptions, it is unclear whether the static label is duplicated three times, whether the alignment module is trained with a different loss, or whether the semantic branch is omitted entirely. The accompanying sentence that 'the original Task-Adapter already accomplishes semantic alignment during feature extraction' does not resolve this, since Task-Adapter is a visual-only module. Please specify the exact configuration of row 3 and, if a static-text duplicate is used, justify it; as written, this row cannot support the claim that fine-grained alignment is redundant without enriched semantics.
  3. [4.1, Tables 1-2] All results are reported as averages over 10,000 episodes without variance or confidence intervals. Several improvements over the closest baselines are small: in Table 2, Task-Adapter++ ties TSAM at 96.2 on Kinetics 1-shot, and several UCF101 margins are around 0.2-0.5%. Without error bars or repeated-run statistics, the headline claim of consistent state-of-the-art performance is not fully supported. Please report mean and standard deviation over at least three runs, or confidence intervals for the 10,000-task estimates, and discuss significance for the small-margin cases.
  4. [3.7, Eq. (18)] Equation (18) multiplies the video matching 'probability' P from Eq. (11) by the video-text matching 'probability' from Eq. (17). Since both quantities are described as probabilities or normalized cosine scores, their product is not a logit; if the product is used directly as the cross-entropy logit, the model is effectively optimizing a different objective than the one described. Please clarify whether P and the video-text score are raw similarity scores, whether the product is re-normalized before loss computation, or whether the intended combination is log P plus the log of the video-text score.
minor comments (7)
  1. [3.3, Eq. (5)] Equation (5) has malformed parentheses; it should be written as z_l = Adapter(LN(z_s_l)) + MLP(LN(z_s_l)) + z_s_l if the adapter is parallel to the MLP as stated in the text.
  2. [3.4] The sentence 'we first map all the videos ... into their patch embeddings by Eq. (1) and Eq. (2)' is imprecise, since Eqs. (1)-(2) describe ViT block updates rather than patch embedding construction.
  3. [3.5] Please specify the exact output shape of the semantic features used in Eq. (17): if the end-of-text tokens of the three sub-action prompts are concatenated, the result should be stated as a C x 3 x D tensor so that the indexing F_T^c[s] is unambiguous.
  4. [4.1] The statement that 'the first three datasets focus on scene understanding' is inaccurate for Kinetics, which is primarily an action dataset; consider saying 'scene-related datasets' or 'datasets with less stringent temporal requirements.'
  5. [Figure 8] Figure 8 has no axis labels or legend; as printed, the reader cannot tell which curve corresponds to which dataset or what the x-axis represents. Please add these details.
  6. [Table 3] The term 'Partial Adapting' is used before it is defined; please define it in the text, for example as inserting adapters only into the last L layers, before the table is discussed.
  7. [Table 2] On Kinetics 1-shot, Task-Adapter++ ties TSAM at 96.2; the claim of consistent state-of-the-art performance should be qualified for tied or statistically indistinguishable cases.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the method is self-contained and benchmarked externally.

full rationale

Task-Adapter++ makes no claim that a derived quantity is predicted from inputs that already contain it. The final score in Eq. 18 is a product of two independently computed matching terms: a visual metric score (Eq. 11) and a cross-modal score (Eq. 17). Neither term is fitted to the test-set labels reported in Tables 1 and 2; both are trained on base-class episodes and evaluated on held-out splits. The LLM-generated sub-action descriptions are fixed before training, as the paper states that 'The semantic corpus is saved as json file without any manual modification,' and they are not optimized against test accuracy, so the cross-modal term is not a fitted input renamed as a prediction. Hyperparameters such as adapter depth and frame count are chosen by validation and are disclosed as such. The only self-citations are the disclosed Task-Adapter MM'24 baseline and ordinary related-work references; these are used for comparison, not as load-bearing evidence that the new adapters work. The unverified assumption that GPT-4's beginning/process/end descriptions line up with the fixed three-segment partition in Eq. 16 is a robustness and correctness concern, not a circularity: Eq. 16 imposes a fixed partition and could hurt accuracy if the assumption fails, which is exactly what an independent empirical check would reveal. Hence no circular step can be exhibited, and the appropriate finding is no significant circularity.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central claim is empirical benchmark performance. The main free parameters are architectural depths and frame count chosen by validation, not fitted to the test set. The key domain assumptions are about transferability of CLIP to video, temporal ordering of LLM descriptions, and the validity of equal or overlapping video segmentation. No genuinely new entities, forces, or conserved quantities are introduced.

free parameters (4)
  • Task-Adapter insertion depth L = 2 or 6 depending on dataset
    Chosen per benchmark on validation accuracy (Fig. 8a, Table 8); scene datasets use top 2 layers, SSv2 uses top 6.
  • Order-Adapter insertion depth M = 2 or 8 depending on dataset
    Chosen per benchmark on validation accuracy (Fig. 8b, Table 8); SSv2 uses top 2, scene datasets use top 8.
  • Number of sampled frames T = 8
    Uniform sampling of 8 frames per video, following CLIP-FSAR and MA-FSAR; this determines the length of the temporal segments in Eq. 16.
  • Cross-attention layers in alignment module = 1
    Ablation in Table 7 shows 1-layer cross-attention works best; deeper variants reduce accuracy.
assumptions (4)
  • domain assumption Frozen CLIP ViT features, augmented with small adapters, transfer to video action recognition and preserve generalization.
    Used throughout Section 3.4; the paper compares PEFT against full fine-tuning in Table 3 and relies on this assumption for the whole framework.
  • domain assumption LLM-generated sub-action descriptions are temporally ordered and correspond to the visual stages of the action.
    Invoked in Section 3.5 when constructing prompts 'A video of action about {LABEL}: {SUB-ACTIONS-1/2/3}' and in the cross-modal matching of Section 3.6.
  • domain assumption Equal division of video frames into three segments approximates the semantic beginning/process/end stages.
    Invoked in Eq. 16, where the 7 aligned frames are averaged into three segments; no evidence is given that the temporal boundaries match the semantic sub-actions.
  • domain assumption The frozen text encoder, with an order adapter after the last M layers, can model the sequential relationship between sub-actions using a reused MSA layer.
    Section 3.5, Eqs. 12 to 14; this is the core mechanism of the semantic branch.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Task-Adapter++: Task-specific Adaptation with Order-aware Alignment for Few-shot Action Recognition." pith.science (2026). https://pith.science/paper/7VGI7AL7

@misc{pith2026250506002,
  author       = {Pith},
  title        = {Pith review of: Task-Adapter++: Task-specific Adaptation with Order-aware Alignment for Few-shot Action Recognition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7VGI7AL7}},
  note         = {Machine review of arXiv:2505.06002}
}
read the original abstract

Large-scale pre-trained models have achieved remarkable success in language and image tasks, leading an increasing number of studies to explore the application of pre-trained image models, such as CLIP, in the domain of few-shot action recognition (FSAR). However, current methods generally suffer from several problems: 1) Direct fine-tuning often undermines the generalization capability of the pre-trained model; 2) The exploration of task-specific information is insufficient in the visual tasks; 3) The semantic order information is typically overlooked during text modeling; 4) Existing cross-modal alignment techniques ignore the temporal coupling of multimodal information. To address these, we propose Task-Adapter++, a parameter-efficient dual adaptation method for both image and text encoders. Specifically, to make full use of the variations across different few-shot learning tasks, we design a task-specific adaptation for the image encoder so that the most discriminative information can be well noticed during feature extraction. Furthermore, we leverage large language models (LLMs) to generate detailed sequential sub-action descriptions for each action class, and introduce semantic order adapters into the text encoder to effectively model the sequential relationships between these sub-actions. Finally, we develop an innovative fine-grained cross-modal alignment strategy that actively maps visual features to reside in the same temporal stage as semantic descriptions. Extensive experiments fully demonstrate the effectiveness and superiority of the proposed method, which achieves state-of-the-art performance on 5 benchmarks consistently. The code is open-sourced at https://github.com/Jaulin-Bage/Task-Adapter-pp.

Figures

Figures reproduced from arXiv: 2505.06002 by the authors.

Figure 1
Figure 1. Comparison of different paradigms. (a) Existing methods fully fine-tune the feature extractor and combine it with temporal alignment module. (b) Existing methods extract semantic features from static enhanced descriptions and calculate the score by global matching. (c) For the image encoder, we freeze the backbone and apply task-specific adaptation to extract the most discriminative information. (d) For the text enc… view at source ↗
Figure 2
Figure 2. Illustration of Task-Adapter++. We first augment the labels with sub-action descriptions using LLMs, and then independently integrate task adapters and semantic order adapters into the visual and se￾mantic branches. After feature extraction, the support and query features are input into a metric module to compute classification scores, while the query features are further processed through a cross-modal align￾ment m… view at source ↗
Figure 3
Figure 3. Illustration of Task-Adapter in visual branch. For task-specific adaptation, we introduce task [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: AIM (a) adapts the standard ViT bock (b) by freezing the original pre-trained model (outlined with [PITH_FULL_IMAGE:figures/full_fig_p012_4.png]
Figure 5
Figure 5. Figure 5: Illustration of order adaptation in semantic branch. For semantic order adaptation, we introduce [PITH_FULL_IMAGE:figures/full_fig_p015_5.png]
Figure 6
Figure 6. Figure 6: The structure of our fine-grained cross-modal alignment. We use cross-attention for fine-grained [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]
Figure 7
Figure 7. Figure 7: The stage-wise illustration of the cross-modal feature matching process. The aligned query features [PITH_FULL_IMAGE:figures/full_fig_p018_7.png]
Figure 8
Figure 8. Figure 8: Effect of inserting Task-Adapter into the last L layers (a) and semantic order adapter (b) into the last M layers (e.g., L, M = 1, 2, 3, ..., 12) on 3 different datasets. The best is highlighted with a black star. The effect of inserting position of adapters. In this p…
Figure 9
Figure 9. Figure 9: The results of N-way 1-shot between different methods on SSv2-Small and Kinetics. N-way 1-shot results [PITH_FULL_IMAGE:figures/full_fig_p025_9.png]
Figure 10
Figure 10. Figure 10: Comparison of the performance achieved by combining di [PITH_FULL_IMAGE:figures/full_fig_p026_10.png]
Figure 11
Figure 11. Figure 11: Visualization of the attention maps and cross-modal matching scores for the “Pouring something [PITH_FULL_IMAGE:figures/full_fig_p029_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

66 extracted references · 52 canonical work pages

  1. [1]

    Radford, J

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, et al., Learning transferable visual models from natural language supervision, in: International conference on machine learning, PMLR, 2021, pp. 8748–8763

    2021

  2. [2]

    J. Li, D. Li, C. Xiong, S. Hoi, Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation, in: International con- ference on machine learning, PMLR, 2022, pp. 12888–12900

    work page 2022

  3. [3]

    P. Gao, S. Geng, R. Zhang, T. Ma, R. Fang, Y . Zhang, H. Li, Y . Qiao, Clip-adapter: Better vision-language models with feature adapters, International Journal of Computer Vision (2023) 1–15

    work page 2023

  4. [4]

    Z. Zhou, Y . Lei, B. Zhang, L. Liu, Y . Liu, Zegclip: Towards adapting clip for zero-shot semantic segmentation, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 11175–11185

    work page 2023

  5. [5]

    T. Yang, Y . Zhu, Y . Xie, A. Zhang, C. Chen, M. Li, Aim: Adapting image models for efficient video action recognition, arXiv preprint arXiv:2302.03024 (2023)

    arXiv 2023

  6. [6]

    Wortsman, G

    M. Wortsman, G. Ilharco, J. W. Kim, M. Li, S. Kornblith, R. Roelofs, R. G. Lopes, H. Hajishirzi, A. Farhadi, H. Namkoong, et al., Robust fine-tuning of zero-shot models, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 7959–7971

    work page 2022

  7. [7]

    Carreira, A

    J. Carreira, A. Zisserman, Quo vadis, action recognition? a new model and the kinetics dataset, in: proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 6299–6308

    2017

  8. [8]

    something 28 something

    R. Goyal, S. Ebrahimi Kahou, V . Michalski, J. Materzynska, S. Westphal, H. Kim, V . Haenel, I. Fruend, P. Yianilos, M. Mueller-Freitag, et al., The" something 28 something" video database for learning and evaluating visual common sense, in: Proceedings of the IEEE international conference on computer vision, 2017, pp. 5842–5850

    work page 2017

Show all 66 references
  1. [9]

    Z. Peng, Z. Li, J. Zhang, Y . Li, G.-J. Qi, J. Tang, Few-shot image recognition with knowledge transfer, in: Proceedings of the IEEE/CVF international conference on computer vision, 2019, pp. 441–449

    2019

  2. [10]

    Chen, Y .-C

    W.-Y . Chen, Y .-C. Liu, Z. Kira, Y .-C. F. Wang, J.-B. Huang, A closer look at few-shot classification, arXiv preprint arXiv:1904.04232 (2019)

    2019 arXiv

  3. [11]

    Y . Tian, Y . Wang, D. Krishnan, J. B. Tenenbaum, P. Isola, Rethinking few-shot image classification: a good embedding is all you need?, in: Computer Vision– ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XIV 16, Springer, 2020, pp. 266–282

    2020

  4. [12]

    H.-J. Ye, H. Hu, D.-C. Zhan, F. Sha, Few-shot learning via embedding adapta- tion with set-to-set functions, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 8808–8817

    2020

  5. [13]

    C. Cao, Y . Zhang, Learning to compare relation: Semantic alignment for few-shot learning, IEEE Transactions on Image Processing 31 (2022) 1462–1474

    2022

  6. [14]

    K. Cao, J. Ji, Z. Cao, C.-Y . Chang, J. C. Niebles, Few-shot video classification via temporal alignment, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 10618–10627

    2020

  7. [15]

    Perrett, A

    T. Perrett, A. Masullo, T. Burghardt, M. Mirmehdi, D. Damen, Temporal- relational crosstransformers for few-shot action recognition, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 475–484

    2021

  8. [16]

    X. Wang, S. Zhang, Z. Qing, M. Tang, Z. Zuo, C. Gao, R. Jin, N. Sang, Hybrid relation guided set matching for few-shot action recognition, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 19948–19957. 29

    2022

  9. [17]

    S. Li, H. Liu, R. Qian, Y . Li, J. See, M. Fei, X. Yu, W. Lin, Ta2n: Two-stage action alignment network for few-shot action recognition, in: Proceedings of the AAAI Conference on Artificial Intelligence, volume 36, 2022, pp. 1404–1411

    2022

  10. [18]

    X. Wang, S. Zhang, H. Yuan, Y . Zhang, C. Gao, D. Zhao, N. Sang, Few- shot action recognition with captioning foundation models, arXiv preprint arXiv:2310.10125 (2023)

    2023 arXiv

  11. [19]

    J. Xing, M. Wang, X. Hou, G. Dai, J. Wang, Y . Liu, Multimodal adaptation of clip for few-shot action recognition, arXiv preprint arXiv:2308.01532 (2023)

    2023 arXiv

  12. [20]

    Y . Yu, C. Cao, Y . Zhang, Q. Lv, L. Min, Y . Zhang, Building a multi-modal spatiotemporal expert for zero-shot action recognition with clip, arXiv preprint arXiv:2412.09895 (2024)

    2024 arXiv

  13. [21]

    B. Li, M. Liu, G. Wang, Y . Yu, Frame order matters: A temporal sequence-aware model for few-shot action recognition, Proceedings of the AAAI Conference on Artificial Intelligence 39 (2025) 18218–18226

    2025

  14. [22]

    Y . Du, F. Liu, L. Jiao, S. Li, Z. Hao, P. Li, J. Wang, H. Wang, X. Liu, Text gener- ation and multi-modal knowledge transfer for few-shot object detection, Pattern Recognition 161 (2025) 111283

    2025

  15. [23]

    Houlsby, A

    N. Houlsby, A. Giurgiu, S. Jastrzebski, B. Morrone, Q. De Laroussilhe, A. Ges- mundo, M. Attariyan, S. Gelly, Parameter-efficient transfer learning for nlp, in: International Conference on Machine Learning, PMLR, 2019, pp. 2790–2799

    2019

  16. [24]

    J. He, C. Zhou, X. Ma, T. Berg-Kirkpatrick, G. Neubig, Towards a unified view of parameter-efficient transfer learning, arXiv preprint arXiv:2110.04366 (2021)

    2021 arXiv

  17. [25]

    Lester, R

    B. Lester, R. Al-Rfou, N. Constant, The power of scale for parameter-efficient prompt tuning, arXiv preprint arXiv:2104.08691 (2021)

    2021 arXiv

  18. [26]

    H. Li, D. Eigen, S. Dodge, M. Zeiler, X. Wang, Finding task-relevant features for few-shot learning by category traversal, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 1–10. 30

    2019

  19. [27]

    H. Liu, W. Lv, J. See, W. Lin, Task-adaptive spatial-temporal video sampler for few-shot action recognition, in: Proceedings of the 30th ACM International Conference on Multimedia, 2022, pp. 6230–6240

    2022

  20. [28]

    C. Cao, Y . Zhang, Y . Yu, Q. Lv, L. Min, Y . Zhang, Task-adapter: Task-specific adaptation of image models for few-shot action recognition, in: Proceedings of the 32nd ACM International Conference on Multimedia, MM ’24, Association for Computing Machinery, New York, NY , USA,...

    2024

  21. [29]

    Mehrotra, A

    A. Mehrotra, A. Dukkipati, Generative adversarial residual pairwise networks for one shot learning, arXiv preprint arXiv:1703.08033 (2017)

    2017 arXiv

  22. [30]

    Y .-X. Wang, R. Girshick, M. Hebert, B. Hariharan, Low-shot learning from imag- inary data, in: Proceedings of the IEEE conference on computer vision and pat- tern recognition, 2018, pp. 7278–7286

    2018

  23. [31]

    Snell, K

    J. Snell, K. Swersky, R. Zemel, Prototypical networks for few-shot learning, Advances in neural information processing systems 30 (2017)

    2017

  24. [32]

    Vinyals, C

    O. Vinyals, C. Blundell, T. Lillicrap, D. Wierstra, et al., Matching networks for one shot learning, Advances in neural information processing systems 29 (2016)

    2016

  25. [33]

    J. Xie, F. Long, J. Lv, Q. Wang, P. Li, Joint distribution matters: Deep brownian distance covariance for few-shot classification, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 7972–7981

    2022

  26. [34]

    M. Dong, F. Li, Z. Li, X. Liu, Prsn: Prototype resynthesis network with cross- image semantic alignment for few-shot image classification, Pattern Recognition 159 (2025) 111122

    2025

  27. [35]

    X. Li, Z. Guo, R. Zhu, Z. Ma, J. Guo, J.-H. Xue, A simple scheme to amplify inter-class discrepancy for improving few-shot fine-grained image classification, Pattern Recognition 156 (2024) 110736

    2024

  28. [36]

    F. Zhou, P. Wang, L. Zhang, W. Wei, Y . Zhang, Meta-collaborative comparison for effective cross-domain few-shot learning, Pattern Recognition 156 (2024) 110790. 31

    2024

  29. [37]

    C. Finn, P. Abbeel, S. Levine, Model-agnostic meta-learning for fast adaptation of deep networks, in: International conference on machine learning, PMLR, 2017, pp. 1126–1135

    2017

  30. [38]

    Xu, J.-F

    J. Xu, J.-F. Ton, H. Kim, A. Kosiorek, Y . W. Teh, Metafun: Meta-learning with iterative functional updates, in: International Conference on Machine Learning, PMLR, 2020, pp. 10617–10627

    2020

  31. [39]

    Requeima, J

    J. Requeima, J. Gordon, J. Bronskill, S. Nowozin, R. E. Turner, Fast and flexible multi-task classification using conditional neural adaptive processes, Advances in Neural Information Processing Systems 32 (2019)

    2019

  32. [40]

    Bateni, R

    P. Bateni, R. Goyal, V . Masrani, F. Wood, L. Sigal, Improved few-shot visual classification, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 14493–14502

    2020

  33. [41]

    Ma, Z.-D

    Z.-X. Ma, Z.-D. Chen, L.-J. Zhao, Z.-C. Zhang, X. Luo, X.-S. Xu, Cross-layer and cross-sample feature optimization network for few-shot fine-grained image classification, in: Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, 2024, pp. 4136–4144

    2024

  34. [42]

    Zhang, Y

    W. Zhang, Y . Zhang, Y . Deng, W. Zhang, J. Lin, B. Huang, J. Zhang, W. Yu, Ta- adapter: Enhancing few-shot clip with task-aware encoders, Pattern Recognition 153 (2024) 110559

    2024

  35. [43]

    Zhang, L

    H. Zhang, L. Zhang, X. Qi, H. Li, P. H. Torr, P. Koniusz, Few-shot action recog- nition with permutation-invariant attention, in: Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part V 16, Springer, 2020, pp. 525–542

    2020

  36. [44]

    X. Zhu, A. Toisoul, J.-M. Perez-Rua, L. Zhang, B. Martinez, T. Xiang, Few- shot action recognition with prototype-centered attentive learning, arXiv preprint arXiv:2101.08085 (2021)

    2021 arXiv

  37. [45]

    Lu, H.-J

    S. Lu, H.-J. Ye, D.-C. Zhan, Few-shot action recognition with compromised metric via optimal transport, arXiv preprint arXiv:2104.03737 (2021). 32

    2021 arXiv

  38. [46]

    Z. Zhu, L. Wang, S. Guo, G. Wu, A closer look at few-shot video classification: A new baseline and benchmark, arXiv preprint arXiv:2110.12358 (2021)

    2021 arXiv

  39. [47]

    Thatipelli, S

    A. Thatipelli, S. Narayan, S. Khan, R. M. Anwer, F. S. Khan, B. Ghanem, Spatio- temporal relation modeling for few-shot action recognition, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 19958–19967

    2022

  40. [48]

    J. Xing, M. Wang, Y . Liu, B. Mu, Revisiting the spatial and temporal modeling for few-shot action recognition, in: Proceedings of the AAAI Conference on Artificial Intelligence, volume 37, 2023, pp. 3001–3009

    2023

  41. [49]

    Zhang, Y

    Y . Zhang, Y . Fu, X. Ma, L. Qi, J. Chen, Z. Wu, Y .-G. Jiang, On the importance of spatial relations for few-shot action recognition, in: Proceedings of the 31st ACM International Conference on Multimedia, 2023, pp. 2243–2251

    2023

  42. [50]

    Y . Fu, L. Zhang, J. Wang, Y . Fu, Y .-G. Jiang, Depth guided adaptive meta-fusion network for few-shot video recognition, in: Proceedings of the 28th ACM Inter- national Conference on Multimedia, 2020, pp. 1142–1151

    2020

  43. [51]

    X. Wang, W. Ye, Z. Qi, X. Zhao, G. Wang, Y . Shan, H. Wang, Semantic-guided relation propagation network for few-shot action recognition, in: Proceedings of the 29th ACM International Conference on Multimedia, 2021, pp. 816–825

    2021

  44. [52]

    Wanyan, X

    Y . Wanyan, X. Yang, C. Chen, C. Xu, Active exploration of multimodal com- plementarity for few-shot action recognition, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 6492–6502

    2023

  45. [53]

    Y . Zhu, Y . Chen, X. Mao, X. Yan, Y . Wang, W. Lu, J. Wang, X. Ji, Enhanc- ing few-shot clip with semantic-aware fine-tuning, IEEE Transactions on Neural Networks and Learning Systems (2024) 1–14

    2024

  46. [54]

    L. Zhu, Y . Yang, Compound memory networks for few-shot video classification, in: Proceedings of the European Conference on Computer Vision (ECCV), 2018, pp. 751–766. 33

    2018

  47. [55]

    Thatipelli, S

    A. Thatipelli, S. Narayan, S. Khan, R. M. Anwer, F. S. Khan, B. Ghanem, Spatio- temporal relation modeling for few-shot action recognition, in: CVPR, 2022

    2022

  48. [56]

    Zheng, S

    S. Zheng, S. Chen, Q. Jin, Few-shot action recognition with hierarchical matching and contrastive learning, in: Computer Vision – ECCV 2022: 17th European Conference, Tel Aviv, Israel, October 23–27, 2022, Proceedings, Part IV , Springer-Verlag, Berlin, Heidelberg, 2...

    2022

  49. [57]

    X. Wang, S. Zhang, Z. Qing, C. Gao, Y . Zhang, D. Zhao, N. Sang, Molo: Motion-augmented long-short contrastive learning for few-shot action recogni- tion, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 18011–18021

    2023

  50. [58]

    X. Wang, S. Zhang, Z. Qing, Z. Zuo, C. Gao, R. Jin, N. Sang, Hyrsm++: Hybrid relation guided temporal set matching for few-shot action recognition, Pattern Recognition 147 (2024) 110110

    2024

  51. [59]

    X. Liu, H. Zhang, H. Pirsiavash, Mastaf: a model-agnostic spatio-temporal at- tention fusion network for few-shot video classification, in: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2023, pp. 2508–2517

    2023

  52. [60]

    X. Wang, S. Zhang, J. Cen, C. Gao, Y . Zhang, D. Zhao, N. Sang, Clip-guided prototype modulating for few-shot action recognition, International Journal of Computer Vision (2023)

    2023

  53. [61]

    Kuehne, H

    H. Kuehne, H. Jhuang, E. Garrote, T. Poggio, T. Serre, Hmdb: A large video database for human motion recognition, International conference on computer vision (2011) 2556–2563

    2011

  54. [62]

    Soomro, A

    K. Soomro, A. R. Zamir, M. Shah, Ucf101: A dataset of 101 human actions classes from videos in the wild, arXiv preprint arXiv:1212.0402 (2012)

    2012 arXiv

  55. [63]

    S. Tu, Q. Dai, Z. Wu, Z.-Q. Cheng, H. Hu, Y .-G. Jiang, Implicit temporal model- ing with learnable alignment for video recognition, in: Proceedings of the ieee/cvf international conference on computer vision, 2023, pp. 19936–19947. 34

    2023

  56. [64]

    L. Zhu, Y . Yang, Label independent memory for semi-supervised few-shot video classification, IEEE Transactions on Pattern Analysis and Machine Intelligence 44 (2020) 273–285

    2020

  57. [65]

    Achiam, S

    OpenAI, J. Achiam, S. Adler, S. Agarwal, L. Ahmad, Gpt-4 technical report, arXiv preprint arXiv:2303.08774 (2023)

    2023 arXiv

  58. [66]

    T. Yu, P. Chen, Y . Dang, R. Huan, R. Liang, Multi-speed global contextual sub- space matching for few-shot action recognition, in: Proceedings of the 31st ACM International Conference on Multimedia, 2023, pp. 2344–2352. 35 Congqi Cao(Member, IEEE) received the B.E. degree in ...

    2023

Pith tools

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