Pith. sign in

REVIEW 4 major objections 5 minor 48 references

A Baseline Study and Benchmark for Few-Shot Open-Set Action Recognition with Feature Residual Discrimination

T0 review · 4 major / 5 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a small discriminator trained on the difference between a query video's features and its nearest class prototype—the Feature-Residual Discriminator—consistently improves rejection of unknown actions across five few-sho

desk verdict A genuinely useful benchmark for few-shot open-set action recognition, but the central FR-Disc comparison is weakened by an unequal training budget on HMDB51/UCF101 and no error bars. read the letter →

arxiv 2603.04125 v2 pith:JXXNEHZT submitted 2026-03-04 cs.CV

Stefano Berti , Giulia Pasquale , Lorenzo Natale This is my paper · ORCID
classification cs.CV
keywords few-shotlearningopen-setrecognitionactionvideounderstandingfeatureresidualprototype-basedclassificationunknownrejectionbenchmark
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 seeks to establish that few-shot action recognition can be made open-set—able to reject videos of actions not present in the small support set—without sacrificing accuracy on known classes. It introduces a benchmark across five video datasets and shows that a lightweight discriminator trained on the feature residual between a query and its nearest class prototype rejects unknown actions more reliably than standard maximum-logit, maximum-softmax, entropy, or garbage-class techniques. The central claim is that this residual signal, rather than raw logit magnitude, best captures the temporal dissimilarity of genuinely unknown actions. If correct, it provides the field with a simple, effective component and a standard evaluation suite for future comparisons.

What carries the argument

The Feature-Residual Discriminator (FR-Disc) is a lightweight auxiliary network that takes the difference between the query video's feature vector and the feature vector of the closest support-class prototype (the class with highest predicted probability). It is trained with a binary cross-entropy loss to output a confidence score indicating whether the query is known or unknown. This residual-based signal is the paper's proposed alternative to using raw logits or softmax scores as open-set confidence, and it is what the authors argue carries the crucial information about whether a query belongs to none of the supported classes.

What would settle it

Train the softmax, entropy, and garbage-class baselines for the same number of iterations as FR-Disc on HMDB51 and UCF101 (removing early stopping), or early-stop FR-Disc to 1K iterations; if the open-set gap disappears or reverses, unequal training budget explains the result. Additionally, test unknown queries drawn from a different dataset or distribution: if FR-Disc's advantage only holds for held-out classes of the same distribution, the 'unknown rejection' claim is weaker than stated.

Watch

Extended reading notes

Core claim

The paper's central discovery is that modeling the discrepancy between query features and the closest support-class prototype with a small binary discriminator yields consistently better known/unknown separation than existing open-set scoring methods. Across five datasets and two few-shot backbones, this Feature-Residual Discriminator improves open-set metrics such as AUROC and OSCR while slightly improving or preserving few-shot classification accuracy. The paper also reports that a well-trained closed-set classifier alone is already a strong open-set detector, with closed-set and open-set performance strongly correlated (Pearson coefficients from 0.84 to 0.99), and that the maximum logit s

Load-bearing premise

The comparison on HMDB51 and UCF101 assumes that giving FR-Disc more training iterations than the logit-based baselines (which needed early stopping to avoid overfitting) does not explain its better rejection; if extra training drives the gain, the residual-discriminator claim collapses.

Editorial extensions

If this is right

  • FR-Disc improves open-set rejection across all five benchmark datasets and on both backbone architectures, while slightly improving few-shot accuracy, making it a reliable default choice for few-shot open-set video recognition.
  • Strong closed-set classifiers already provide competitive open-set rejection; this suggests that improving closed-set feature learning is itself a viable path toward open-set robustness.
  • The maximum logit score consistently beats the maximum softmax score for video data, indicating that preserving logit magnitude is important for uncertainty estimation in spatio-temporal models.
  • Entropic open-set training offers a simple and effective trade-off, particularly for weaker backbones, improving open-set metrics at a minor cost to closed-set accuracy.
  • The five-dataset benchmark provides a standardized protocol for comparing future few-shot open-set action recognition methods.

Reading between the lines

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

  • The residual between query and nearest prototype is a cheap signal that could be combined with logit-based scores (e.g., as an additional feature) rather than replacing them, potentially improving calibration further.
  • The paper's 'unknown' classes are held-out classes from the same dataset distribution; real-world unknowns may be out-of-distribution in style or domain. Testing FR-Disc with cross-dataset unknowns would verify whether the residual signal generalizes beyond label-set mismatch.
  • FR-Disc's training appears to act as a regularizer, producing more compact feature clusters (as shown in t-SNE visualizations); this suggests it might improve closed-set generalization even when open-set rejection is not the primary goal.
  • If the residual-discrimination idea transfers, it could be applied to other few-shot recognition tasks—such as object detection or medical video analysis—where rejecting novel categories is safety-critical.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper introduces a benchmark and baseline study for few-shot open-set action recognition (FSOS-AR), adapting five video datasets and four open-set techniques applied to two few-shot action recognition models, SAFSAR and STRM. The main proposed method, FR-Disc, is a feature-residual discriminator that is trained jointly with the closed-set classifier to accept/reject queries based on the residual between the query feature and the projection of the closest support class. The paper reports that FR-Disc consistently improves open-set metrics over Softmax, EOS, and Garbage Class baselines while preserving or improving closed-set accuracy, and claims a new state-of-the-art for FSOS-AR. The paper also analyzes the MLS/MSS scoring choice, the correlation between closed-set and open-set performance, and provides qualitative visualizations.

Significance. If the central claim is correct, the paper would provide a useful benchmark and a simple, effective method for FSOS-AR, an underexplored problem. The adaptation of five standard datasets and two strong FS-AR backbones, along with the implementation of multiple open-set techniques, is a valuable contribution to the community. The paper also correctly identifies that a strong closed-set classifier is often a strong open-set baseline, echoing results from image-based open-set recognition. The FR-Disc method is clearly described and the code/website are promised, which would aid reproducibility. However, the central claim that FR-Disc 'significantly enhances' unknown rejection is currently under-supported because of a training-protocol asymmetry on two of five datasets and the absence of variance statistics.

major comments (4)
  1. [§6, first paragraph; Table 1] For SAFSAR on HMDB51 and UCF101, the Softmax/EOS/GC baselines are trained for only 1K iterations because of overfitting, while FR-Disc is trained without early stopping. This is a systematic training-budget confound. For example, Table 1 shows HMDB51 1-shot AUROC 77.48 vs. 70.76 and OSCR 68.79 vs. 62.23. The reported gains could be due to longer/stabilized training rather than the residual discriminator mechanism. The paper states that FR-Disc 'did not require early stopping,' but no ablation separates the effect of the discriminator at inference from its effect as a regularizer that enables extended training. To support the superiority claim, the authors should equalize training budgets, e.g. train all methods for the same number of iterations, or early-stop FR-Disc on these datasets, and show the gains persist.
  2. [Tables 1 and 2; §6.2] All metrics in Tables 1 and 2 are reported as single numbers with no standard deviations or confidence intervals over seeds or test episodes. The text repeatedly uses 'significantly' and 'consistently' to describe FR-Disc's gains, but without variance estimates these claims are not statistically supported. For a benchmark paper, it is standard to report mean±std over multiple random seeds and episodic splits. This is especially important for small gains, e.g. UCF101 1-shot OSCR 89.01 vs. 88.31 for SAFSAR, or STRM HMDB51 OSCR 69.11 vs. 67.34, where the differences may be within run-to-run noise.
  3. [§4.2, Eq. (5)–(7); §6] The mechanism underlying FR-Disc is not fully isolated. The proposed method changes both the training objective (joint loss with L_Disc) and the inference score (discriminator output). The observed improvement on HMDB51/UCF101 could come from the auxiliary loss acting as an implicit regularizer rather than from the residual-based discrimination at inference. The authors should include an ablation that evaluates the Softmax baseline with the same joint loss but with MLS/MSS at inference, or evaluate FR-Disc with the closed-set model early-stopped at 1K iterations. Without such an ablation, the contribution of the residual discriminator per se is not established.
  4. [Abstract; §6.2] The claim of setting a 'new state-of-the-art for FSOS-AR' is not supported by the evidence reported. The comparison is only against internal baselines (Softmax, EOS, GC) on a benchmark introduced in the same paper; no existing FSOS-AR methods from the literature are compared. Since the authors state that the only prior FSOS-AR work is their own skeleton-based method, the 'state-of-the-art' claim is self-referential and should be either removed or clearly framed as 'state-of-the-art within the proposed benchmark.'
minor comments (5)
  1. [§6.2] The symbol for the EOS loss weight is inconsistently written: αEOS in §4.2 but σEOS in §6.2. Please unify.
  2. [Eq. (6)] The notation φ(SS_i, argmax_j({p̂_ij})) is ambiguous. The residual should be defined as the difference between the query feature and the prototype of the predicted support class. Please clarify what φ(SS_i, j) denotes and explicitly state how prototypes are computed.
  3. [§6.1] The scoring baseline is not identical across models: MLS is used for SAFSAR and MSS for STRM because of logit boundedness. This is a reasonable pragmatic choice, but it should be stated more prominently so readers do not compare raw numbers across the two tables as if the baseline definition were the same.
  4. [Figure 2] The textual labels 'NoY es' and the flow diagram are hard to read. Please improve the figure's typography and clarify the decision paths for the three open-set techniques.
  5. [§5.1] The random 70/30 splits for NTURGBD and Diving48 are described, but the random seed is not reported. For reproducibility, specify the seed or state that multiple seeds are used and results averaged.

Circularity Check

0 steps flagged · score 0.0 of 10

No derivation-level circularity; FR-Disc is evaluated against independent baselines on external datasets.

full rationale

This paper is an empirical benchmark study rather than a formal derivation, and no claimed result reduces to its inputs by construction. The FR-Disc open-set score is a learned function of feature residuals (Eq. 5-7) trained with ground-truth known/unknown labels; it is not defined in terms of the evaluation metrics or fitted to test outcomes. The baselines (Softmax, EOS, GC) are independent configurations. The benchmark uses pre-existing external datasets (HMDB51, UCF101, SSv2, NTURGBD, Diving48) with standard episodic splits. Self-citations to [2] identify the origin of the FR-Disc concept but do not carry the empirical claim, which is established by fresh experiments in Tables 1-2. The paper explicitly discloses unequal training budgets for SAFSAR on HMDB51/UCF101 (§6, first paragraph), where baselines were limited to 1K iterations while FR-Disc was not early-stopped; this is a possible confound for the comparison, but it is a fairness/validity concern, not a definitional or self-citational circularity. No equation is equivalent to its own output, and no fitted parameter is renamed as a prediction. Therefore, the circularity score is 0.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central empirical claims rest on several hand-chosen hyperparameters (α_EOS, α_Disc, τ, training budget) and on three domain assumptions about how 'unknown' should be defined and measured. No new physical or conceptual entities are introduced; FR-Disc is a neural module rather than an invented entity.

free parameters (4)
  • α_EOS (entropy loss weight)
    Balances L_CE and L_EOS in the EOS method; the value is not reported in the paper, making exact replication difficult.
  • α_Disc (discriminator loss weight)
    Balances L_CE and L_Disc for FR-Disc; the value is not reported in the paper.
  • Threshold τ for OS ACC = 0.5
    Fixed to 0.5 in §5.2 for all methods and datasets. This is an arbitrary choice that interacts with the balanced known/unknown evaluation set.
  • Training budget for SAFSAR baselines = 1K iterations on HMDB51 and UCF101
    Chosen by hand in §6 to avoid overfitting for Softmax/EOS/GC, while FR-Disc is trained longer. This ad hoc choice creates an uncontrolled variable in the comparison.
assumptions (3)
  • domain assumption Unknown test classes are disjoint from support classes but drawn from the same dataset distribution as known classes.
    The open-set protocol in §3.1 and the splits in §5.1 define unknowns as held-out classes from the same dataset. If real-world unknowns come from a different distribution, the results may not transfer.
  • ad hoc to paper The closed-set model's correct/incorrect prediction is a valid binary label for training the FR-Disc discriminator.
    In Eq. 5, positive discriminator samples are defined as known queries correctly classified by the closed-set model (I(ŷ=y)). This couples the discriminator to the base classifier's mistakes and may bias the learned known/unknown boundary.
  • domain assumption Balanced known/unknown query sets are the appropriate evaluation distribution.
    OS ACC and AUROC in §3.2 are computed over a balanced set T_known ∪ T_unknown. Real deployment priors may be heavily imbalanced, so threshold-dependent OS ACC at τ=0.5 may not reflect operational use.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Baseline Study and Benchmark for Few-Shot Open-Set Action Recognition with Feature Residual Discrimination." pith.science (2026). https://pith.science/paper/JXXNEHZT

@misc{pith2026260304125,
  author       = {Pith},
  title        = {Pith review of: A Baseline Study and Benchmark for Few-Shot Open-Set Action Recognition with Feature Residual Discrimination},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JXXNEHZT}},
  note         = {Machine review of arXiv:2603.04125}
}
read the original abstract

Few-Shot Action Recognition (FS-AR) has shown promising results but is often limited by a closed-set assumption that fails in real-world open-set scenarios. While Few-Shot Open-Set (FSOS) recognition is well-established for images, its extension to spatio-temporal video data remains underexplored. To address this, we propose an architectural extension based on a Feature-Residual Discriminator (FR-Disc), adapting previous work on skeletal data to the more complex video domain. Extensive experiments on five datasets demonstrate that while common open-set techniques provide only marginal gains, our FR-Disc significantly enhances unknown rejection capabilities without compromising closed-set accuracy, setting a new state-of-the-art for FSOS-AR. The project website, code, and benchmark are available at: https://hsp-iit.github.io/fsosar/.

Figures

Figures reproduced from arXiv: 2603.04125 by the authors.

Figure 1
Figure 1. Comparison between the results of the closed-set Softmax Base￾line SAFSAR [34] and its proposed open-set extension FR-Disc on an unknown query. The Support Set contains K = 2 classes and N = 1 videos. The true class of the query is not contained in the Support Set, so it should be rejected through low confidence (e.g., < 50%). The Baseline incorrectly classifies it as known (false positive), while FR-Disc correctly … view at source ↗
Figure 2
Figure 2. Overview of closed- and open-set prediction flows for (a) implicit (Softmax, [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Analysis of open-set scoring and performance correlation. (a) Comparison [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison between the SAFSAR Softmax baseline and FR [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

48 extracted references · 6 linked inside Pith

  1. [1]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Bao, W., Yu, Q., Kong, Y.: Evidential deep learning for open set action recognition. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 13349–13358 (2021)

    2021

  2. [2]

    In: 2022 IEEE-RAS 21st International Conference on Humanoid Robots (Humanoids)

    Berti, S., Rosasco, A., Colledanchise, M., Natale, L.: One-shot open-set skeleton- based action recognition. In: 2022 IEEE-RAS 21st International Conference on Humanoid Robots (Humanoids). pp. 765–772. IEEE (2022)

    2022

  3. [3]

    arXiv preprint arXiv:2406.09112 (2024)

    Bisgin, H., Palechor, A., Suter, M., Günther, M.: Large-scale evaluation of open-set image classification techniques. arXiv preprint arXiv:2406.09112 (2024)

    arXiv 2024

  4. [4]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Cao, K., Ji, J., Cao, Z., Chang, C.Y., Niebles, J.C.: Few-shot video classification via temporal alignment. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 10618–10627 (2020)

    2020

  5. [5]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Cen, J., Zhang, S., Wang, X., Pei, Y., Qing, Z., Zhang, Y., Chen, Q.: Enlarging instance-specific and class-specific information for open-set action recognition. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 15295–15304 (2023)

    2023

  6. [6]

    IEEE Transactions on Pattern Analysis and Machine Intelligence44(11), 8065–8081 (2021)

    Chen, G., Peng, P., Wang, X., Tian, Y.: Adversarial reciprocal points learning for open set recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence44(11), 8065–8081 (2021)

    2021

  7. [7]

    arXiv preprint arXiv:2405.16146 (2024)

    Chen, X., Li, Y., Chen, H.: Dual-adapter: Training-free dual adaptation for few-shot out-of-distribution detection. arXiv preprint arXiv:2405.16146 (2024)

    arXiv 2024

  8. [8]

    IEEE Transactions on Multimedia 25, 5763–5774 (2022)

    Deng, S., Yu, J.G., Wu, Z., Gao, H., Li, Y., Yang, Y.: Learning relative feature displacement for few-shot open-set recognition. IEEE Transactions on Multimedia 25, 5763–5774 (2022)

    2022

Show all 48 references
  1. [9]

    Advances in Neural Information Processing Systems31(2018)

    Dhamija, A.R., Günther, M., Boult, T.: Reducing network agnostophobia. Advances in Neural Information Processing Systems31(2018)

    2018

  2. [10]

    G. Koch, R. Zemel, R. Salakhutdinov: Siamese Neural Networks for One-shot Image Recognition. In: ICMLw (2015)

    2015

  3. [11]

    arXiv preprint arXiv:1707.07418 (2017)

    Ge, Z., Demyanov, S., Chen, Z., Garnavi, R.: Generative openmax for multi-class open set classification. arXiv preprint arXiv:1707.07418 (2017)

    2017 arXiv

  4. [12]

    something something

    Goyal, R., Ebrahimi Kahou, S., Michalski, V., Materzynska, J., Westphal, S., Kim, H., Haenel, V., Fruend, I., Yianilos, P., Mueller-Freitag, M., et al.: The" something something" video database for learning and evaluating visual common sense. In: Proceedings of the IEEE intern...

    2017

  5. [13]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Huang, S., Ma, J., Han, G., Chang, S.F.: Task-adaptive negative envision for few-shot open-set recognition. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 7171–7180 (2022)

    2022

  6. [14]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Jeong, M., Choi, S., Kim, C.: Few-shot open-set recognition by transformation consistency. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 12566–12575 (2021)

    2021

  7. [15]

    arXiv preprint arXiv:1705.06950 (2017)

    Kay, W., Carreira, J., Simonyan, K., Zhang, B., Hillier, C., Vijayanarasimhan, S., Viola, F., Green, T., Back, T., Natsev, P., et al.: The kinetics human action video dataset. arXiv preprint arXiv:1705.06950 (2017)

    2017 arXiv

  8. [16]

    In: 2011 International conference on computer vision

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

    2011

  9. [17]

    In: Proceedings of the European Conference on Computer Vision (ECCV)

    Li, Y., Li, Y., Vasconcelos, N.: Resound: Towards action recognition without representation bias. In: Proceedings of the European Conference on Computer Vision (ECCV). pp. 513–528 (2018) 14 S. Berti et al

    2018

  10. [18]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Liu, B., Kang, H., Li, H., Hua, G., Vasconcelos, N.: Few-shot open-set recognition using meta-learning. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 8798–8807 (2020)

    2020

  11. [19]

    Journal of machine learning research9(Nov), 2579–2605 (2008)

    Maaten, L.v.d., Hinton, G.: Visualizing data using t-sne. Journal of machine learning research9(Nov), 2579–2605 (2008)

    2008

  12. [20]

    In: European Conference on Computer Vision

    Miller, D., Sünderhauf, N., Kenna, A., Mason, K.: Open-set recognition in the age of vision-language models. In: European Conference on Computer Vision. pp. 1–18. Springer (2024)

    2024

  13. [21]

    In: Proceedings of the European conference on computer vision (ECCV)

    Neal, L., Olson, M., Fern, X., Wong, W.K., Li, F.: Open set learning with coun- terfactual images. In: Proceedings of the European conference on computer vision (ECCV). pp. 613–628 (2018)

    2018

  14. [22]

    arXiv preprint arXiv:2408.02052 (2024)

    Ochal, M., Patacchiola, M., Boudiaf, M., Wang, S.: Eol: Transductive few-shot open-set recognition by enhancing outlier logits. arXiv preprint arXiv:2408.02052 (2024)

    2024 arXiv

  15. [23]

    Artificial Intelligence Review54(3), 2259–2322 (2021)

    Pareek, P., Thakkar, A.: A survey on video-based human action recognition: recent updates, datasets, challenges, and applications. Artificial Intelligence Review54(3), 2259–2322 (2021)

    2021

  16. [24]

    In: Proceedings of the AAAI conference on artificial intelligence

    Peng, K., Yin, C., Zheng, J., Liu, R., Schneider, D., Zhang, J., Yang, K., Sarfraz, M.S., Stiefelhagen, R., Roitberg, A.: Navigating open set scenarios for skeleton- based action recognition. In: Proceedings of the AAAI conference on artificial intelligence. vol. 38, pp. 4487–...

    2024

  17. [25]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

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

    2021

  18. [26]

    Neural Networks176, 106326 (2024)

    Qin, Y., Liu, B.: Tfrs: A task-level feature rectification and separation method for few-shot video action recognition. Neural Networks176, 106326 (2024)

    2024

  19. [27]

    Roady, R., Hayes, T.L., Kemker, R., Gonzales, A., Kanan, C.: Are open set classifi- cation methods effective on large-scale datasets? Plos one15(9), e0238302 (2020)

    2020

  20. [28]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Shahroudy, A., Liu, J., Ng, T.T., Wang, G.: Ntu rgb+ d: A large scale dataset for 3d human activity analysis. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 1010–1019 (2016)

    2016

  21. [29]

    arXiv preprint arXiv:2211.12030 (2022)

    Shi, Y., Wu, X., Lin, H.: Knowledge prompting for few-shot action recognition. arXiv preprint arXiv:2211.12030 (2022)

    2022 arXiv

  22. [30]

    IEEE Transactions on Multimedia (2024)

    Shi, Y., Wu, X., Lin, H., Luo, J.: Commonsense knowledge prompting for few-shot action recognition in videos. IEEE Transactions on Multimedia (2024)

    2024

  23. [31]

    Song, N., Zhang, C., Lin, G.: Few-shot open-set recognition using background as unknowns.In:Proceedingsofthe30thACMInternationalConferenceonMultimedia. pp. 5970–5979 (2022)

    2022

  24. [32]

    IEEE Transactions on Image Processing33, 1389–1402 (2024)

    Su, B., Zhang, H., Li, J., Zhou, Z.: Toward generalized few-shot open-set object detection. IEEE Transactions on Image Processing33, 1389–1402 (2024)

    2024

  25. [33]

    Pattern Recognition151, 110400 (2024)

    Sun, L.Y., Chu, W.T.: Overall positive prototype for few-shot open-set recognition. Pattern Recognition151, 110400 (2024)

    2024

  26. [34]

    In: Proceedings of the IEEE/CVF Winter Conference on Appli- cations of Computer Vision

    Tang, Y., Béjar, B., Vidal, R.: Semantic-aware video representation for few-shot action recognition. In: Proceedings of the IEEE/CVF Winter Conference on Appli- cations of Computer Vision. pp. 6458–6468 (2024)

    2024

  27. [35]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

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

    2022

  28. [36]

    Advances in neural information processing systems35, 10078–10093 (2022)

    Tong, Z., Song, Y., Wang, J., Wang, L.: Videomae: Masked autoencoders are data-efficient learners for self-supervised video pre-training. Advances in neural information processing systems35, 10078–10093 (2022)

    2022

  29. [37]

    Vaze, S., Han, K., Vedaldi, A., Zisserman, A.: Open-set recognition: A good closed- set classifier is all you need? (2021)

    2021

  30. [38]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Wang, H., Pang, G., Wang, P., Zhang, L., Wei, W., Zhang, Y.: Glocal energy- based learning for few-shot open-set recognition. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 7507–7516 (2023)

    2023

  31. [39]

    International Journal of Computer Vision132(6), 1899–1912 (2024)

    Wang, X., Zhang, S., Cen, J., Gao, C., Zhang, Y., Zhao, D., Sang, N.: Clip-guided prototype modulating for few-shot action recognition. International Journal of Computer Vision132(6), 1899–1912 (2024)

    1912

  32. [40]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

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

    2023

  33. [41]

    Pattern Recognition147, 110110 (2024)

    Wang, X., Zhang, S., Qing, Z., Zuo, Z., Gao, C., Jin, R., Sang, N.: Hyrsm++: Hybrid relation guided temporal set matching for few-shot action recognition. Pattern Recognition147, 110110 (2024)

    2024

  34. [42]

    IEEE Transactions on Image Processing (2024)

    Wang, X., Yan, Y., Hu, H.M., Li, B., Wang, H.: Cross-modal contrastive learning network for few-shot action recognition. IEEE Transactions on Image Processing (2024)

    2024

  35. [43]

    In: Proceedings of the 29th ACM International Conference on Multimedia

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

    2021

  36. [44]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

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

    2023

  37. [45]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Wu, J., Zhang, T., Zhang, Z., Wu, F., Zhang, Y.: Motion-modulated temporal fragment alignment network for few-shot action recognition. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 9151–9160 (2022)

    2022

  38. [46]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Xing, J., Wang, M., Ruan, Y., Chen, B., Guo, Y., Mu, B., Dai, G., Wang, J., Liu, Y.: Boosting few-shot action recognition with graph-guided hybrid matching. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 1740–1750 (2023)

    2023

  39. [47]

    IEEE Transactions on Neural Networks and Learning Systems (2024)

    Yin, X., Cao, B., Hu, Q., Wang, Q.: Rd-openmax: Rethinking openmax for robust realistic open-set recognition. IEEE Transactions on Neural Networks and Learning Systems (2024)

    2024

  40. [48]

    In: Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part V 16

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

    2020

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.