Pith. sign in

REVIEW 4 major objections 6 minor 70 references

3DPyranet Features Fusion for Spatio-temporal Feature Learning

T0 review · 4 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read A four-layer 3D pyramidal network with position-oriented, partially shared weights is claimed to learn video features that rival or beat far larger 3D convolutional networks, reaching 94.83% on Maryland and 98.99% on Weizmann with only…

desk verdict A legitimate compact 3D pyramidal architecture, but the reported state-of-the-art results are undercut by conflicting baseline numbers and non-equivalent evaluation protocols. read the letter →

arxiv 2504.18977 v1 pith:ZLWDHQDW submitted 2025-04-26 cs.CV

Ihsan Ullah , Alfredo Petrosino This is my paper · ORCID
classification cs.CV
keywords 3Dpyramidalneuralnetworkspatio-temporalfeaturelearninghumanactionrecognitiondynamicscenefusionlinearSVMvideoclassificationparameter-efficientdeep
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 tries to establish that a deliberately shallow, biologically inspired pyramid architecture can learn video features that compete with much larger deep 3D convolutional networks. It proposes 3DPyraNet, a 3D correlation network with partially shared, position-oriented weights that process several adjacent frames, and 3DPyraNet-F, a version that fuses the top-layer feature maps and classifies them with a linear SVM. The reported result is that 3DPyraNet-F reaches 98.99% on Weizmann, 93.42% on KTH, 93.67% on YUPENN, and 94.83% on Maryland, using roughly 0.83M parameters where C3D uses 17.5M. If true, the claim matters because it would show that network shape and weight-sharing philosophy, not depth and width alone, can drive accuracy on small labeled video sets, and that a small model can be deployed where memory is scarce.

What carries the argument

The load-bearing object is the 3D partially shared weight matrix: for every output neuron, three weight slices and three consecutive input frames are correlated through a receptive field of size RF, with the weight matrix the same size as the input so that the kernel is position-specific and locally shared only to the degree set by an overlap O. This weighting scheme does the work of learning spatial layout and temporal correspondence at once; 3D temporal max pooling then selects maxima across three fields, and the final normalized feature maps are concatenated into one vector that a linear SVM classifies.

What would settle it

Run 3DPyraNet-F on KTH and Weizmann under the exact standard protocols used by the cited baselines, using fixed public splits, multiple runs, and per-class standard deviations. If the mean accuracy falls below the reported 93.42% on KTH or 98.99% on Weizmann by more than a couple of points, the state-of-the-art comparison is not robust to split choice.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that replacing the sliding convolutional kernel with a 3D, partially shared, position-oriented weight matrix—each output neuron gets a unique local 3D kernel assembled from three consecutive frames—lets a four-layer network learn discriminative spatio-temporal features from raw or lightly preprocessed frames. Training that base model and then extracting and fusing the final normalized feature maps into a single vector for a linear SVM yields 3DPyraNet-F. The paper reports that this fusion step raises accuracy on Weizmann from 90.9% to 98.99%, and on Maryland from 67% to 94.83%, overtaking C3D by 7.17 percentage points there; on KTH and YUPENN it reports results comparable or close to recent deep baselines while keeping parameters below one million.

Load-bearing premise

The headline numbers rest on the assumption that the paper's loosely described training and evaluation splits match the baselines it is compared with, so the reported margins are not an artifact of easier or harder test sets.

Editorial extensions

If this is right

  • A shallow 3D pyramidal model can match or beat much deeper 3D convolutional networks on small action and scene datasets, so depth is not the only route to discriminative video features.
  • Fusing the highest-layer learned features and classifying with a linear SVM is itself a performance lever: the paper reports Weizmann rising from 90.9% to 98.99% with this addition.
  • The roughly 0.83M-parameter model, compared with C3D's 17.5M, implies video recognition models can run in embedded or memory-constrained settings if the accuracy carries over.
  • The strong Maryland result suggests position-oriented temporal correlation is especially suited to camera-induced motion, where the scene content shifts between frames.

Reading between the lines

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

  • If the protocol issue were resolved, a natural extension would be to train 3DPyraNet-F on a large-scale video corpus; the paper itself suggests this only as future work, and the outcome is not known.
  • The gap between global fusion (3DPyraNet-F) and local mean fusion (3DPyraNet-F_M) on Weizmann and KTH hints that fusion granularity is a tunable axis; sweeping fusion over intermediate layers rather than only the top layer could lift YUPENN, where the model lags the best baseline by 5.33%.
  • The position-oriented weight matrix may be interpretable by construction: visualizing the trained weight matrices, as the paper reports for the first layer, could provide a direct test of which spatial positions and temporal offsets the model relies on for each class.
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 manuscript proposes 3DPyraNet, a three-hidden-layer 3D pyramidal neural network whose partially shared position-oriented weight scheme is intended to learn spatio-temporal features from video, and 3DPyraNet-F, which fuses the last-layer features and classifies them with a linear SVM. Experiments are reported on Weizmann, KTH, YUPENN, and Maryland for action and dynamic scene recognition. The paper reports 98.99% on Weizmann, 93.42% on KTH, 93.67% on YUPENN, and 94.83% on Maryland for 3DPyraNet-F, with 0.83M parameters versus 17.5M for C3D, and claims state-of-the-art performance on three of the four datasets. Back-propagation equations for the 3D pyramidal, pooling, and fully connected layers are given in Section 3.3.

Significance. The architectural idea of a biologically motivated pyramidal 3D weight-sharing scheme is coherent, and the parameter budget (0.83M) relative to C3D is attractive if the accuracies can be reproduced. The derivation of the back-propagation equations in Section 3.3 is one of the more concrete parts of the paper. However, the empirical contribution is the main selling point, and it is currently underdetermined by the evaluation protocol: split definitions, sampling procedures, standard deviations, and clip-level versus video-level metrics are missing, and several baseline numbers in Tables 2-4 conflict. The SOTA margins are therefore not reproducible as stated.

major comments (4)
  1. [Section 5.2, Table 4 (KTH)] The KTH evaluation is not protocol-equivalent to the cited baselines. The text states that the model was tested on a 'random set of samples from full KTH datasets' with only background subtraction, whereas Baccouche et al.'s 3D-ConvNet baseline is evaluated on the KTH1/KTH2 complexity split with a voting scheme, and Ji et al.'s 3DCNN uses ROI sequences extracted by a tracker. The reported 93.42% for 3DPyraNet-F and the 89.40%/90.2% baseline values therefore do not measure the same quantity. Please specify the exact split, class-wise composition, number of runs, standard deviations, and whether results are clip-level or video-level; without these details the KTH rows of Table 4 are not comparable.
  2. [Section 5.3.2 and Tables 2/4 (Maryland)] The Maryland state-of-the-art claim is internally inconsistent. The text says 3DPyraNet-F reaches 94.83% and outperforms C3D by 7.17%; that margin is computed from the 87.7% C3D figure in Table 4, but Table 2 reports C3D's Maryland overall accuracy as 78 and 3DPyraNet-F's as 95. The same inconsistency occurs for YUPENN, where Table 3 gives C3D overall 97 and Table 4 gives 98.1. Please reconcile the baseline values and recompute all claimed gains, or state explicitly which C3D configuration each table refers to.
  3. [Section 5.2 and Table 4 (Weizmann)] The Weizmann protocol is not specified. The paper reports a mean accuracy over five splits but does not define the splits, the number of training/testing sequences per class, or the standard deviation. It also states that the 92.46% 'all-1' result is shown as 3DPyraNet(all-1) in Table 4, but Table 4 contains no such entry. Please provide the split definition, per-split results, and a corrected table or reference.
  4. [Section 5.3.2 (YUPENN comparisons)] The margin statements for YUPENN are mutually inconsistent. The text says the model falls short of state-of-the-art by 5.33% and later says Feichtenhofer et al. (2014) is better by 1.5%, while Table 4 lists 99 and 96.2 for the two Feichtenhofer entries and 93.67 for 3DPyraNet-F, corresponding to differences of 5.33 and 2.53 percentage points. Please correct the text and table so that the claimed margins are reproducible.
minor comments (6)
  1. [Section 3.2] The notation table is referred to as 'Table. ??' in two places, but no notation table appears; please add it or remove the references.
  2. [Section 5.1] The abbreviations 'AR' and 'SR' are used, with 'SR' presumably meaning dynamic scene recognition (DSR); this abbreviation is never defined.
  3. [Table 4 and Section 4.2] The model name appears as '3DPyraNet-FM' in Table 4 and as '3DPyraNet-F M' in the text; please use one consistent name throughout.
  4. [Section 5.2] The sentence 'it overcame reported best result by 3DConvNet model, i.e. 88.26% with an average of 91.07% from ten tests' is ambiguous about which model produced the 91.07% average; please rewrite for clarity.
  5. [Equations 1-2] The symbol 'D' is used both for temporal depth and as the upper limit in summations; a consistent table of symbols would help the reader.
  6. [General] There are typographical errors such as 'becuase', 'hihg', and inconsistent capitalization; a careful proofread is needed.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the empirical claims rest on held-out SVM evaluation, not on inputs defined in terms of outputs.

full rationale

The paper's central contribution is an architecture plus a fusion/SVM pipeline. The accuracy numbers are obtained by training 3DPyraNet, extracting Norm6 features, training a linear SVM on those features, and classifying test-set feature vectors; Section 4.1 states 'SVM trains over these fused features. Finally, the trained SVM model is used to classify the feature vectors extracted from the testing set using 3DPyraNet.' This is an external held-out evaluation, not a quantity defined by the fitted parameters themselves. The architecture derivation is explicit: Eq. (1) defines the 3D correlation output, Eq. (2) defines the receptive field, Eq. (3) defines pooling, and Eqs. (5)-(26) give backpropagation updates; none of these equations contains the target accuracies as inputs. The only self-citations are to prior pyramidal-network work (Cantoni and Petrosino 2002; Ullah and Petrosino 2016) as architectural inspiration and as a general remark about parameter reduction; no load-bearing uniqueness theorem or ansatz is imported from those citations, and no benchmark number is derived from them. Concerns raised by the reader about non-equivalent evaluation protocols and inconsistent baseline numbers across Tables 2-4 are verifiability/correctness issues, not circularity: they affect whether the reported comparisons are meaningful, but they do not make the reported accuracies equal to the inputs of the method by construction.

Assumptions & free parameters 5 free parameters · 3 assumptions · 0 invented entities

The paper introduces no new physical or mathematical entities. The free parameters are standard architectural hyperparameters, but several, especially RF, O, and D, are explicitly tuned for performance and the central accuracy claims depend on them. The key assumption is that the reported experimental protocol is fair and reproducible, which the text does not fully support.

free parameters (5)
  • RF (receptive field size) = 4 or 3 depending on layer and dataset
    RF controls the correlation window size in 3DCORR layers. Values are chosen by hand for each dataset and layer, and the paper states 'RF and O are tuned for handling the performance.'
  • O (overlap value) = 3, 2, or 0 depending on layer and dataset
    O controls how much neighboring output neurons share weights. The paper explicitly says RF and O are the main tunable parameters for performance.
  • Temporal depth D = 3
    D sets the number of consecutive frames in the 3D correlation kernel. The paper states 'D is 3 as shown in Table ??', and this value is chosen as a design decision.
  • Number of weight matrices per layer = 3
    The paper says 'we used three 3D weight matrix at first layer' and keeps the set of kernels fixed across layers. This is an architecture choice directly affecting model capacity.
  • Learning rate schedules = 0.00015 with decay for AR; 0.000015 with decay for DSR
    Learning rate and decay schedule are chosen per dataset and are tuned for convergence. These values affect final accuracy but are not derived from any principle.
assumptions (3)
  • domain assumption The proposed 3D correlation weighting scheme improves feature learning for video recognition compared to standard convolutions.
    The paper assumes that partially shared, position-oriented weights are more discriminative than fully shared kernels for actions and dynamic scenes. This is asserted based on prior work and qualitative motivation, not proven.
  • domain assumption Accuracy numbers from the paper's experimental runs are representative and were obtained without test-set peeking or cherry-picking.
    The paper reports mean accuracies and per-class tables, but provides insufficient protocol detail to verify that the reported numbers are not selected from favorable runs.
  • domain assumption The compared baselines used equivalent evaluation protocols.
    The paper compares against numbers taken from other papers without checking whether train/test splits, class subsets, or clip-level versus video-level metrics match. This is explicitly questionable for KTH, where the authors use a random subset.

how reviews work

0 comments
Cite this review

Pith. "Pith review of 3DPyranet Features Fusion for Spatio-temporal Feature Learning." pith.science (2026). https://pith.science/paper/ZLWDHQDW

@misc{pith2026250418977,
  author       = {Pith},
  title        = {Pith review of: 3DPyranet Features Fusion for Spatio-temporal Feature Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZLWDHQDW}},
  note         = {Machine review of arXiv:2504.18977}
}
read the original abstract

Convolutional neural network (CNN) slides a kernel over the whole image to produce an output map. This kernel scheme reduces the number of parameters with respect to a fully connected neural network (NN). While CNN has proven to be an effective model in recognition of handwritten characters and traffic signal sign boards, etc. recently, its deep variants have proven to be effective in similar as well as more challenging applications like object, scene and action recognition. Deep CNN add more layers and kernels to the classical CNN, increasing the number of parameters, and partly reducing the main advantage of CNN which is less parameters. In this paper, a 3D pyramidal neural network called 3DPyraNet and a discriminative approach for spatio-temporal feature learning based on it, called 3DPyraNet-F, are proposed. 3DPyraNet introduces a new weighting scheme which learns features from both spatial and temporal dimensions analyzing multiple adjacent frames and keeping a biological plausible structure. It keeps the spatial topology of the input image and presents fewer parameters and lower computational and memory costs compared to both fully connected NNs and recent deep CNNs. 3DPyraNet-F extract the features maps of the highest layer of the learned network, fuse them in a single vector, and provide it as input in such a way to a linear-SVM classifier that enhances the recognition of human actions and dynamic scenes from the videos. Encouraging results are reported with 3DPyraNet in real-world environments, especially in the presence of camera induced motion. Further, 3DPyraNet-F clearly outperforms the state-of-the-art on three benchmark datasets and shows comparable result for the fourth.

Figures

Figures reproduced from arXiv: 2504.18977 by the authors.

Figure 1
Figure 1. Types of weighting schemes / receptive fields (increase features), hence, as a result the number of parameters increases as compared to PyraNet. The LC kernels in PyraNet are partially shared with another neuron (based on overlap value) as shown with same color con￾nection in [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Weighted scheme for 2D vs our 3D Scheme [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Proposed 3DPyraNet (other than SVM). With SVM it becomes [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Samples from KTH (1 st row) and Weizmann (2 nd row) datasets [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: Samples from YUPENN (1 st row) and MaryLand (2 nd row) datasets proven to be more challenging as compared to static cameras. YUPENN (static camera) consists of 420 videos (fixed size) of 14 scene categories (listed in Table.3). Similarly, MaryLand (non-static camera) c…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

70 extracted references · 51 canonical work pages

  1. [1]

    Baccouche, F

    M. Baccouche, F. Mamalet, C. Wolf, C. Garcia, and A. Baskurt. Sequential deep learning for human action recognition. In Proceedings of the Second International Conference on Human Behavior Unterstanding, HBU'11, pages 29--39, Berlin, Heidelberg, 2011. Springer-Verlag

    work page 2011

  2. [2]

    Ballan, M

    L. Ballan, M. Bertini, A. Del Bimbo , L. Seidenari, and G. Serra. Effective codebooks for human action representation and classification in unconstrained videos . IEEE Transactions on Multimedia, 14 0 (4 PART 2): 0 1234--1245, 2012. ISSN 15209210. doi:10.1109/TMM.2012.2191268

    work page arXiv 2012

  3. [3]

    W. Beil. Volume image processing (vip'93) steerable filters and invariance theory. Pattern Recognition Letters, 15 0 (5): 0 453 -- 460, 1994. ISSN 0167-8655. doi:http://dx.doi.org/10.1016/0167-8655(94)90136-8

    work page doi:10.1016/0167-8655(94)90136-8 1994

  4. [4]

    Y. Bengio. Practical recommendations for gradient-based training of deep architectures. In Neural Networks: Tricks of the Trade, pages 437--478. Springer, 2012

    work page 2012

  5. [5]

    Blank, L

    M. Blank, L. Gorelick, E. Shechtman, M. Irani, and R. Basri. Actions as space-time shapes . In Tenth IEEE International Conference on Computer Vision Volume 1, pages 1395--1402 Vol. 2, 2005

    work page 2005

  6. [6]

    Burt and E

    P. Burt and E. Adelson. The laplacian pyramid as a compact image code. IEEE Transactions on Communications, 31 0 (4): 0 532--540, Apr 1983. ISSN 0090-6778. doi:10.1109/TCOM.1983.1095851

    arXiv 1983

  7. [7]

    Cantoni and A

    V. Cantoni and A. Petrosino. Neural recognition in a pyramidal structure . IEEE Transactions on Neural Networks, 13 0 (2): 0 472--480, 2002. doi:1045-9277(02)01806-4

    work page 2002

  8. [8]

    L.-C. Chen, G. Papandreou, I. Kokkinos, K. Murphy, and A. L. Yuille. Semantic image segmentation with deep convolutional nets and fully connected crfs. ICLR, 2015

    work page 2015

Show all 70 references
  1. [9]

    K. G. Derpanis, M. Lecce, K. Daniilidis, and R. P. Wildes. Dynamic scene understanding: The role of orientation features in space and time in scene classification . IEEE Conference on CVPR, pages 1306--1313, 2012. ISSN 1063-6919

    2012

  2. [10]

    Doll\' a r, V

    P. Doll\' a r, V. Rabaud, G. Cottrell, and S. Belongie. Behavior recognition via sparse spatio-temporal features . Proceedings - 2nd Joint IEEE International Workshop on Visual Surveillance and Performance Evaluation of Tracking and Surveillance, VS-PETS, 2005: 0 65--72, 2005

    2005

  3. [11]

    A. A. Efros, A. C. Berg, G. Mori, and J. Malik. Recognizing action at a distance . IEEE International Conference on Computer Vision, 2003. ISSN 1478-7814. doi:10.1109/ICCV.2003.1238420

    2003 arXiv

  4. [12]

    Feichtenhofer, A

    C. Feichtenhofer, A. Pinz, and R. P. Wildes. Spacetime forests with complementary features for dynamic scene recognition. In BMVC, page 6. Citeseer, 2013

    2013

  5. [13]

    Feichtenhofer, A

    C. Feichtenhofer, A. Pinz, and R. Wildes. Bags of spacetime energies for dynamic scene recognition. In Proceedings of the IEEE Conference on CVPR, pages 2681--2688, 2014

    2014

  6. [14]

    Feichtenhofer, A

    C. Feichtenhofer, A. Pinz, and R. Wildes. Dynamic Scene Recognition with Complementary Spatiotemporal Features . IEEE Transactions on PAMI, PP 0 (99): 0 1, 2016. ISSN 0162-8828. doi:10.1109/TPAMI.2016.2526008

    2016

  7. [15]

    B. J. Fernandes, G. D. Cavalcanti, and T. I. Ren. A receptive field based approach for face detection. In 2009 International Joint Conference on Neural Networks, pages 803--810. IEEE, 2009 a

    2009

  8. [16]

    B. J. T. Fernandes, G. D. C. Cavalcanti, and T. I. Ren. Nonclassical Receptive Field Inhibition Applied to Image Segmentation Receptive and Inhibitory Fields . Neural Network World, 19 0 (1): 0 21--37, 2009 b

    2009

  9. [17]

    B. J. T. Fernandes, G. D. C. Cavalcanti, and T. I. Ren. Lateral inhibition pyramidal neural network for image classification. IEEE transactions on cybernetics, 43 0 (6): 0 2082--91, dec 2013. ISSN 2168-2275. doi:10.1109/TCYB.2013.2240295

    2013

  10. [18]

    N. D. Freitas. Deep learning of invariant spatio-temporal features from video. In Workshop on Deep Learning and Unsupervised Feature Learning in NIPS, pages 1--9, 2010

    2010

  11. [19]

    Fukushima

    K. Fukushima. Neocognitron: A hierarchical neural network capable of visual pattern recognition . Neural Networks, 1 0 (2): 0 119--130, Jan. 1988. ISSN 08936080

    1988

  12. [20]

    Glorot and Y

    X. Glorot and Y. Bengio. Understanding the difficulty of training deep feedforward neural networks. In Proceedings of the thirteenth international conference on artificial intelligence and statistics, pages 249--256. JMLR Workshop and Conference Proceedings, 2010

    2010

  13. [21]

    Goodfellow, Y

    I. Goodfellow, Y. Bengio, and A. Courville. Deep learning. Book in preparation for MIT Press, 2016. URL http://www.deeplearningbook.org

    2016

  14. [22]

    B. Graham. Fractional max-pooling. CoRR, abs/1412.6071, 2014

    2014 arXiv

  15. [23]

    S. Han, H. Mao, and W. J. Dally. A deep neural network compression pipeline: Pruning, quantization, huffman encoding. arXiv preprint arXiv:1510.00149, 10, 2015 a

    2015 arXiv

  16. [24]

    S. Han, J. Pool, J. Tran, and W. J. Dally. Learning both weights and connections for efficient neural networks. CoRR, abs/1506.02626, 2015 b

    2015 arXiv

  17. [25]

    K. He, X. Zhang, S. Ren, and J. Sun. Delving deep into rectifiers: Surpassing human-level performance on imagenet classification. In IEEE International Conference on Computer Vision, pages 1026--1034, 2015

    2015

  18. [26]

    K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770--778, 2016

    2016

  19. [27]

    Huang, Z

    G. Huang, Z. Liu, L. Van Der Maaten, and K. Q. Weinberger. Densely connected convolutional networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 4700--4708, 2017

    2017

  20. [28]

    S. Ji, M. Yang, and K. Yu. 3D convolutional neural networks for human action recognition. IEEE transactions on PAMI, 35 0 (1): 0 221--31, 2013. ISSN 1939-3539

    2013

  21. [29]

    Karpathy and T

    A. Karpathy and T. Leung. Large-scale Video Classification with Convolutional Neural Networks . Proceedings of 2014 IEEE Conference on CVPR, pages 1725--1732, 2014

    2014

  22. [30]

    Krizhevsky, I

    A. Krizhevsky, I. Sutskever, and G. E. Hinton. ImageNet Classification with Deep Convolutional Neural Networks . In Advances in Neural Information Processing Systems, pages 1097--1105, 2012

    2012

  23. [31]

    Laptev, B

    I. Laptev, B. Caputo, C. Sch\" u ldt, and T. Lindeberg. Local velocity-adapted motion events for spatio-temporal recognition. Comput. Vis. Image Underst., 108 0 (3): 0 207--229, Dec. 2007. ISSN 1077-3142

    2007

  24. [32]

    Lazebnik and C

    S. Lazebnik and C. Schmid. Beyond Bags of Features : Spatial Pyramid Matching for Recognizing Natural Scene Categories . Proceedings of the IEEE Computer Society Conference on CVPR, 2: 0 2169--2178, 2006. doi:10.1109/CVPR.2006.68

    2006 doi

  25. [33]

    Q. V. Le, W. Y. Zou, S. Y. Yeung, and A. Y. Ng. Learning hierarchical invariant spatio-temporal features for action recognition with independent subspace analysis . Proceedings of the IEEE Computer Society Conference on CVPR, pages 3361--3368, 2011

    2011

  26. [34]

    Lecun, L

    Y. Lecun, L. Bottou, Y. Bengio, and P. Haffner. Gradient-based learning applied to document recognition . Proceedings of the IEEE, 86 0 (11): 0 2278--2324, 1998

    1998

  27. [35]

    C.-Y. Lee, S. Xie, P. Gallagher, Z. Zhang, and Z. Tu. Deeply-supervised nets. In Artificial intelligence and statistics, pages 562--570. Pmlr, 2015

    2015

  28. [36]

    M. Lin, Q. Chen, and S. Yan. Network in network. CoRR, abs/1312.4400, 2013

    2013 arXiv

  29. [37]

    W. Liu, Z. Wang, D. Tao, and J. Yu. Hessian Regularized Sparse Coding for Human Action Recognition . In 21st International Conference on MMM, pages 502--511, 2015

    2015

  30. [38]

    J. Long, E. Shelhamer, and T. Darrell. Fully convolutional networks for semantic segmentation . Cvpr 2015, 2015

    2015

  31. [39]

    D. G. Lowe. Distinctive image features from scale-invariant keypoints. International Journal of Computer Vision, 60 0 (2): 0 91--110, 2004. ISSN 1573-1405. doi:10.1023/B:VISI.0000029664.99615.94

    2004

  32. [40]

    A. L. Maas, A. Y. Hannun, A. Y. Ng, et al. Rectifier nonlinearities improve neural network acoustic models. In Proc. icml, volume 30, page 3. Atlanta, GA, 2013

    2013

  33. [41]

    Maddalena and A

    L. Maddalena and A. Petrosino. The 3dsobs+ algorithm for moving object detection. Computer Vision and Image Understanding, 122: 0 65 -- 73, 2014

    2014

  34. [42]

    Maninis, P

    K. Maninis, P. Koutras, and P. Maragos. Advances on action recognition in videos using an interest point detector based on multiband spatio-temporal energies. In 2014 IEEE International Conference on Image Processing (ICIP), pages 1490--1494, 2014. doi:10.1109/ICIP.2014.7025298

    2014

  35. [43]

    Melfi, S

    R. Melfi, S. Kondra, and A. Petrosino. Human activity modeling by spatio temporal textural appearance . Pattern Recognition Letters, 34 0 (15): 0 1990--1994, Nov. 2013

    1990

  36. [44]

    G. B. Orr and K.-R. M \"u ller. Neural networks: tricks of the trade. Springer, 2003

    2003

  37. [45]

    Y. Pang, M. Sun, X. Jiang, and X. Li. Convolution in convolution for network in network. IEEE transactions on neural networks and learning systems, 29 0 (5): 0 1587--1597, 2017

    2017

  38. [46]

    S. L. Phung and A. Bouzerdoum. A pyramidal neural network for visual pattern recognition. IEEE transactions on neural networks / a publication of the IEEE Neural Networks Council, 18 0 (2): 0 329--43, Mar. 2007

    2007

  39. [47]

    Schindler and L

    K. Schindler and L. Van Gool . Action Snippets: How many frames does human action recognition require? 26th IEEE Conference on CVPR, 2008

    2008

  40. [48]

    Sch\" u ldt, I

    C. Sch\" u ldt, I. Laptev, and B. Caputo. Recognizing human actions: A local SVM approach . Proceedings - International Conference on Pattern Recognition, 3: 0 32--36, 2004. ISSN 10514651. doi:10.1109/ICPR.2004.1334462

    2004 arXiv

  41. [49]

    Scovanner, S

    P. Scovanner, S. Ali, and M. Shah. A 3-dimensional sift descriptor and its application to action recognition . Proceedings of the ACM International Conference on Multimedia (MM 2007), page 357, 2007. doi:10.1145/1291233.1291311

    2007

  42. [50]

    Simonyan and A

    K. Simonyan and A. Zisserman. Very deep convolutional networks for large-scale image recognition. CoRR, abs/1409.1556, 2014

    2014 arXiv

  43. [51]

    A. F. Smeaton, P. Over, and W. Kraaij. High- L evel F eature D etection from V ideo in TRECV id: a 5- Y ear R etrospective of A chievements. In A. Divakaran, editor, Multimedia Content Analysis, Theory and Applications, pages 151--174. Springer Verlag, Berlin, 2009. ISBN 978-0...

    2009

  44. [52]

    Szegedy, W

    C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhoucke, and A. Rabinovich. Going deeper with convolutions. In CVPR , USA, June 7-12 , pages 1--9, 2015

    2015

  45. [53]

    Taigman, M

    Y. Taigman, M. Yang, M. Ranzato, and L. Wolf. DeepFace: Closing the Gap to Human-Level Performance in Face Verification . In IEEE Conference on CVPR, pages 1701--1708. IEEE, jun 2014. ISBN 978-1-4799-5118-5. doi:10.1109/CVPR.2014.220

    2014 doi

  46. [54]

    G. W. Taylor, R. Fergus, Y. LeCun, and C. Bregler. Convolutional learning of spatio-temporal features . Lecture Notes in Computer Science, 6316 LNCS 0 (PART 6): 0 140--153, 2010. ISSN 03029743

    2010

  47. [55]

    Theriault, N

    C. Theriault, N. Thome, and M. Cord. Dynamic scene classification: Learning motion descriptors with slow features analysis. In IEEE Conference on CVPR, pages 2603--2610, June 2013 a

    2013

  48. [56]

    Theriault, N

    C. Theriault, N. Thome, and M. Cord. Dynamic scene classification: Learning motion descriptors with slow features analysis . Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, pages 2603--2610, 2013 b . ISSN 10636919

    2013

  49. [57]

    D. Tran, L. Bourdev, R. Fergus, L. Torresani, and M. Paluri. Learning Spatiotemporal Features with 3D Convolutional Networks . International Conference on Computer Vision, 2015

    2015

  50. [58]

    Uetz and S

    R. Uetz and S. Behnke. Locally-connected hierarchical neural networks for gpu-accelerated object recognition. In NIPS 2009 Workshop on Large-Scale Machine Learning: Parallelism and Massive Datasets, volume 10, page 13, 2009

    2009

  51. [59]

    Ullah and A

    I. Ullah and A. Petrosino. About pyramid structure in convolutional neural networks. In International Joint Conference on Neural Networks (IJCNN), pages 1318--1324, 2016. doi:10.1109/IJCNN.2016.7727350

    2016

  52. [60]

    H. Wang, M. M. Ullah, A. Klaser, I. Laptev, and C. Schmid. Evaluation of local spatio-temporal features for action recognition . British Machine Vision Conference, pages 124.1--124.11, 2009

    2009

  53. [61]

    H. Wang, A. Kläser, C. Schmid, and C.-L. Liu. Action recognition by dense trajectories. In CVPR 2011, pages 3169--3176, 2011. doi:10.1109/CVPR.2011.5995407

    2011

  54. [62]

    P. Wang, Y. Cao, C. Shen, L. Liu, and H. Shen. Temporal pyramid pooling based convolutional neural network for action recognition. IEEE Transactions on Circuits and Systems for Video Technology, 2016

    2016

  55. [63]

    Weinland, M

    D. Weinland, M. \" O zuysal, and P. Fua. Making action recognition robust to occlusions and viewpoint changes . Lecture Notes in Computer Science, 6313 LNCS 0 (PART 3): 0 635--648, 2010

    2010

  56. [64]

    Yang and Y

    X. Yang and Y. Tian. Action Recognition Using Super Sparse Coding Vector with Spatio-temporal Awareness . In ECCV, volume 8690, pages 727--741, 2014. ISBN 978-3-319-10604-5

    2014

  57. [65]

    Yangqing Jia , Chang Huang , and T. Darrell. Beyond spatial pyramids: Receptive field learning for pooled image features . In IEEE Conference on CVPR, pages 3370--3377. IEEE, jun 2012. ISBN 978-1-4673-1228-8. doi:10.1109/CVPR.2012.6248076

    2012

  58. [66]

    M. D. Zeiler and R. Fergus. Stochastic pooling for regularization of deep convolutional neural networks. arXiv preprint arXiv:1301.3557, 2013

    2013 arXiv

  59. [67]

    M. D. Zeiler and R. Fergus. Visualizing and understanding convolutional networks. In 13th European Conference on Computer Vision, Zurich, Switzerland, September 6-12, pages 818--833, 2014

    2014

  60. [68]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

  61. [69]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

  62. [70]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.