Pith. sign in

REVIEW 3 major objections 6 minor 45 references

Parse Graph-Based Visual-Language Interaction for Human Pose Estimation

T0 review · 3 major / 6 minor · reviewed 2026-08-04 · deepseek-v4-flash

Pith's one-line read This paper claims that fusing visual and language features through hierarchical parse graphs, with high-level nodes guiding updates of low-level nodes, preserves occluded-joint responses and improves pose estimation accuracy on human and an

desk verdict Modest gains and a plausible fusion module, but the parse-graph hierarchy is asserted, not demonstrated. read the letter →

arxiv 2509.07385 v1 pith:JX2ITWX5 submitted 2025-09-09 cs.CV

classification cs.CV
keywords humanposeestimationparsegraphsvision-languagefusionGuidedModuletop-downdecompositionbottom-upcompositionocclusionhandlingvisual-languagealignment
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 proposes PGVL, a visual-language fusion method for human pose estimation built on parse graphs of feature maps. Instead of fusing entire feature maps globally, it splits visual and language tokens along channels into hierarchical trees, lets corresponding low-level nodes attend across modalities, and uses high-level nodes to guide the updated low-level features. The aim is to keep occluded body parts locally responsive while still using global context to infer invisible joints. The paper reports consistent gains over global-fusion and parse-graph baselines, including 91.7 PCKh@0.5 on MPII with a ViT-B backbone, and shows the same fusion improves animal pose estimation.

What carries the argument

The parse graph of a feature map: visual and language tokens are split along channels into a tree, with the root at level n and leaves at level 0, controlled by a parameter G where each g_k is the number of children a node at level k produces. Low levels focus on local features; high levels on global context. The Guided Module (GM) is the load-bearing part: after children receive cross-attention and context information, they are concatenated, and the parent node and root node each query that concatenation with attention; the two query results are summed and replace the parent feature, so high-semantic nodes guide low-semantic updates. Repeating leaf-to-root produces new fused tokens for both

What would settle it

Run a controlled comparison that keeps total parameters and compute fixed and replaces the hand-chosen channel split G with several random channel partitions of equal sizes; if random partitions reproduce the MPII val gains, the hierarchy itself is not the cause. Or take a per-joint breakdown on OCHuman: if occluded joints gain no more than visible joints, the occlusion-specific claim is not supported.

Watch

Extended reading notes

Core claim

PGVL's central claim is that existing visual-language fusion for pose estimation fails under occlusion because global integration dilutes the weak responses of occluded regions and misaligns joints. The remedy is to decompose each modality's feature map along channels into a parse graph whose low-level nodes retain local detail and whose high-level nodes carry global context. Cross-attention runs between corresponding nodes of the vision and language graphs from leaf to root, and a Guided Module (GM) lets root and parent nodes query the concatenated children, so high-semantic nodes steer the updates of low-semantic nodes that already received cross-modal information. The paper reports that t

Load-bearing premise

The method assumes that splitting a feature map along channels into groups produces nodes that genuinely carry local detail versus global context; if the split is arbitrary, the parse graph is just a partition and the gains could come from added cross-attention capacity rather than from hierarchical parsing.

Editorial extensions

If this is right

  • Occluded human joints can be localized from local visual responses instead of being washed out by global fusion; the reported MPII 91.7 and OCHuman 62.8 results support this.
  • Directional language descriptions carry usable signal: removing left/right words from joint prompts costs 0.2 PCKh on MPII.
  • The same hierarchical fusion transfers to animals: AP-10K reaches 82.1 MAP and Animal-Pose 79.3 MAP, beating prompt-based contrastive learning by 4.3 and 5.0 points.
  • PGVL works as a plug-in fusion module: adding it to an existing multi-person method gains +0.3 AP and +0.6 AP50 on OCHuman, and to a top-down animal method gains +1.0 AP on AP-10K.
  • Each component contributes to the reported result: removing the Guided Module costs 0.4 PCKh, removing cross-attention 0.2, and removing context relations 0.1.

Reading between the lines

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

  • If the channel split G is semantically meaningful, a reader could test it directly by replacing the hand-chosen split with random equal-size channel partitions under identical parameter counts; equal results would suggest the gains come from added attention capacity rather than hierarchy.
  • The occlusion-specific story implies gains should concentrate on occluded joints; a per-joint breakdown on OCHuman or a synthetic occlusion benchmark would show whether the 0.4 PCKh ablation gain from GM is driven by hard cases.
  • The level-wise alignment visualizations suggest PGVL could serve as a weak correspondence mechanism, for example generating pseudo-labels for invisible joints in unlabeled images.
  • If language priors are the active ingredient, then automatically generated or dataset-specific descriptions should work as well as hand-written templates, which would widen the method to settings without curated prompts.
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

3 major / 6 minor

Summary. This paper proposes PGVL, a parse-graph-based visual-language interaction module for human pose estimation. Visual and language tokens are decomposed along channels according to a hand-chosen tree structure G into hierarchical node features; context relations within each modality and cross-modal attention between corresponding nodes are computed level by level, and a Guided Module (GM) uses higher-level nodes to filter/update lower-level nodes that have undergone cross-attention. The processed tokens are used with a visual-language matching loss (VLML) and a standard heatmap regression head. The method is evaluated on MPII, CrowdPose, OCHuman, AP-10K, and Animal-Pose, reporting gains over ViTPose, HRNet, RMPG, PGBS, LAMP, and CLAMP, plus ablations showing contributions from context, cross-attention, GM, and directional prompts.

Significance. If the empirical claims hold, the work provides a consistent, plug-in capable multimodal fusion module with modest gains on five datasets and a clear ablation breakdown. The paper's strengths include the controlled plug-in experiments in Table 7, where adding PGVL improves both LAMP and CLAMP, and the identification of the Guided Module as the largest single contributor. The main weaknesses are that the paper's stated mechanism—channel groups as a semantic hierarchy—is not verified, the experimental protocol lacks variance reporting, and some benchmark comparisons mix several design changes at once. The contribution is incremental but potentially publishable if these issues are addressed.

major comments (3)
  1. [§3.3, Eqs. (4)–(10)] The semantic hierarchy that motivates the method is asserted, not supported. Because every node has spatial size S and attention is applied over the full spatial dimension at all levels (Eqs. 6–10), splitting along channels does not create nodes that 'focus on local features' in any spatial sense. The claim that low-level nodes 'maximize maintenance of responses in occluded areas' is therefore not structurally implied. Please provide evidence that the channel groups correspond to distinct semantic levels (e.g., per-group spatial attention maps, channel probing, or a comparison against random channel partitions with the same G and parameter count). If the groups are arbitrary, the parse-graph interpretation should be revised or softened.
  2. [§4.4, Tables 8–9] Ablation differences are small (0.1–0.4 PCKh) and no error bars or repeated runs are reported. On the MPII val set such differences can be within training variance, and Table 9 shows the final configuration (G={2,2,2}, D={512,256}) was selected after checking several variants. Report mean±std over at least three seeds for the main ablations and for the final configuration, and discuss how validation-based selection of D/G affects the reported benchmark numbers.
  3. [§4.3, Tables 4–5 and Table 7] The headline comparisons to CLAMP mix multiple changes simultaneously: visual-backbone pretraining (ImageNet vs CLIP), network architecture, and PGVL. The cleanest isolation of PGVL is Table 7, which is valuable, but the abstract and Section 4.3 emphasize gains of 4.3 and 5.0 MAP against CLAMP without controlling for these factors. Add same-backbone and same-pretraining comparisons in the main tables, or explicitly state that the gains are for the full network rather than for PGVL alone. Also specify the backbones/pretraining of RMPG and PGBS in Table 1 to make that comparison interpretable.
minor comments (6)
  1. [Abstract] Grammar errors should be corrected: 'It ensuring effective fusion' and 'Next stage.' are incomplete sentences.
  2. [§3.3, Eqs. (6)–(7)] It is not specified how the output of Attention(Concat(T), Concat(T), Concat(T)) is split back into per-node context features C. Please clarify the reshaping/slicing operation.
  3. [Table 9] The table layout is confusing: the row 'Fuse only' appears to span several columns, and the 'Cross-Attention' comparison row should be clearly separated from the PGVL variants.
  4. [§4.2 vs §4.4] Section 4.2 defines the channel-semantic-space set as D, but Section 4.4 refers to 'C={512}'. Use a single notation consistently.
  5. [Figure 5] The caption says the panels show 'in sequence' but the sequence of levels (1, 2, 3) is not clearly labeled in the figure itself. Please add explicit level labels.
  6. [References] Cross-attention in Eqs. (9)–(10) is the standard Transformer attention [27]; citing [14] (ViLBERT) for the operation is imprecise.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: PGVL is fully specified in Equations 4-15 and tested by ablations against a no-PGVL baseline; self-citations are motivational/baseline only.

full rationale

The paper's central claim is that PGVL improves pose estimation via parse-graph-guided visual-language fusion. The PGVL module is defined self-contained in Section 3.3: Eqs. 4-5 define channel decomposition into nodes; Eqs. 6-8 define context self-attention; Eqs. 9-10 define cross-modal attention; Eqs. 11-15 define the Guided Module as an attention query operation over child, parent, and root features. No equation uses the target PCKh/MAP metric as an input, and no fitted parameter is renamed as a prediction. Ablations (Table 8) compare the full model against removing context relations, cross-attention, and GM, and against a 'Base (w/o PGVL)' baseline; Table 9 varies hyperparameters G and D, which are architectural choices rather than fitted outputs. The parse-graph-of-feature-map idea is credited to the authors' earlier RMPG [7] and PGBS [8], but those citations are used for motivation and as baselines; the mechanism is re-derived in this paper and tested on MPII, CrowdPose, OCHuman, AP-10K, and Animal-Pose, so the result does not reduce to a self-citation. The skeptical concern that the channel split's semantic hierarchy is asserted rather than demonstrated is a correctness/verifiability issue, not circularity: even if the hierarchical labels were unjustified, the architecture and ablations remain independent of the claimed conclusions. No specific circular reduction can be quoted from the paper.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The central claim rests on the choice of G and D hyperparameters, the assumption that channel-split features preserve semantics, and the CLIP shared embedding space. No physical entities are introduced.

free parameters (2)
  • G (parse graph decomposition structure) = {2,2,2} for most experiments; {2,2} and {2,2,2} variants in Table 9
    Hand-chosen hyperparameter controlling how many child nodes each node splits into at each level. The paper searches over a few values in Table 9 and picks the best.
  • D (channel numbers for semantic spaces) = {512} default; {512,256} for AP-10K and Animal-Pose; variants in Table 9
    Hand-chosen hyperparameter for the MLP projection into semantic spaces. The paper increases D for animal benchmarks and finds gains.
assumptions (3)
  • domain assumption Parse graph energy model (Zhu & Mumford) provides a valid conceptual basis for the hierarchical decomposition used in the method.
    The paper cites Eq. 1-3 from [6] but does not optimize the energy function; it borrows the vocabulary. Section 3.1.
  • domain assumption CLIP text and image encoders provide a shared embedding space in which joint-name prompts and image patches are comparable.
    Used throughout; the method relies on the pretrained CLIP representations without fine-tuning them, so the alignment quality is assumed.
  • ad hoc to paper Splitting a feature map along channels into a balanced tree (controlled by G) yields nodes that correspond to semantically meaningful local and global parts.
    Section 3.3 states the decomposition is controlled by G. No evidence is given that the channel groups correspond to body parts or semantics; this is the load-bearing modeling choice.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Parse Graph-Based Visual-Language Interaction for Human Pose Estimation." pith.science (2026). https://pith.science/paper/JX2ITWX5

@misc{pith2026250907385,
  author       = {Pith},
  title        = {Pith review of: Parse Graph-Based Visual-Language Interaction for Human Pose Estimation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JX2ITWX5}},
  note         = {Machine review of arXiv:2509.07385}
}
read the original abstract

Parse graphs boost human pose estimation (HPE) by integrating context and hierarchies, yet prior work mostly focuses on single modality modeling, ignoring the potential of multimodal fusion. Notably, language offers rich HPE priors like spatial relations for occluded scenes, but existing visual-language fusion via global feature integration weakens occluded region responses and causes alignment and location failures. To address this issue, we propose Parse Graph-based Visual-Language interaction (PGVL) with a core novel Guided Module (GM). In PGVL, low-level nodes focus on local features, maximizing the maintenance of responses in occluded areas and high-level nodes integrate global features to infer occluded or invisible parts. GM enables high semantic nodes to guide the feature update of low semantic nodes that have undergone cross attention. It ensuring effective fusion of diverse information. PGVL includes top-down decomposition and bottom-up composition. In the first stage, modality specific parse graphs are constructed. Next stage. recursive bidirectional cross-attention is used, purified by GM. We also design network based on PGVL. The PGVL and our network is validated on major pose estimation datasets. We will release the code soon.

Figures

Figures reproduced from arXiv: 2509.07385 by the authors.

Figure 1
Figure 1. PGVL includes in two stages. The top-down decomposition stage converts visual and language [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Our network. Images and text are encoded through their respective encoders to get their own to [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. The specific design of PGVL: CA denotes Cross Attention and GM denotes Guided Module. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: The specific design of Guided Module. The child nodes with cross-attention and context informa [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 3
Figure 3. Figure 3: Next, visual and language results in multiple semantic spaces are concatenated [PITH_FULL_IMAGE:figures/full_fig_p010_3.png]
Figure 5
Figure 5. Figure 5: Examples (a) and (b) demonstrate the hierarchical visual-language alignment results of the PGVL [PITH_FULL_IMAGE:figures/full_fig_p018_5.png]
Figure 6
Figure 6. Figure 6: The vertical orange dashed line demarcates: (left) multi-person pose estimation comparisons be [PITH_FULL_IMAGE:figures/full_fig_p018_6.png]
Figure 7
Figure 7. Figure 7: Single-person results comparing the ablated version (without GM) against our full network on [PITH_FULL_IMAGE:figures/full_fig_p019_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

45 extracted references · 42 canonical work pages

  1. [1]

    S. Du, Z. Zhang, T. Ikenaga, Anatpose: Bidirectionally learning anatomy-aware heatmaps for human pose estimation, Pattern Recogn. 155 (2024) 110654

  2. [2]

    C. Wang, F. Zhang, X. Zhu, S. S. Ge, Low-resolution human pose estimation, Pattern Recogn. 126 (2022) 108579

  3. [3]

    L. Tian, P. Wang, G. Liang, C. Shen, An adversarial human pose estimation net- work injected with graph structure, Pattern Recogn. 115 (2021) 107863

  4. [4]

    Bin, Z.-M

    Y . Bin, Z.-M. Chen, X.-S. Wei, X. Chen, C. Gao, N. Sang, Structure-aware human pose estimation with graph convolutional networks, Pattern Recogn. 106 (2020) 107410

  5. [5]

    Z. Wu, Y . Ding, L. Wan, T. Li, F. Nian, Local and global self-attention en- hanced graph convolutional network for skeleton-based action recognition, Pat- tern Recogn. 159 (2025) 111106. 20

  6. [6]

    S.-C. Zhu, D. Mumford, et al., A stochastic grammar of images, Found. Trends® Comput. Graph. Vision (FTCGV) 2 (4) (2007) 259–362

  7. [7]

    Liu, Shibang and Xie, Xuemei and Shi, Guangming, Refinement module based on parse graph of feature map for human pose estimation, arXiv preprint arXiv:2501.11069 (2025)

  8. [8]

    S. Liu, X. Xie, G. Shi, Human pose estimation via parse graph of body structure, IEEE Trans. Circuits Syst. Video Technol. (2024)

Show all 45 references
  1. [9]

    Lupyan, D

    G. Lupyan, D. H. Rakison, J. L. McClelland, Language is not just for talking: Redundant labels facilitate learning of novel categories, Psychol. Sci. 18 (12) (2007) 1077–1083

  2. [10]

    Boutonnet, B

    B. Boutonnet, B. Dering, N. Viñas-Guasch, G. Thierry, Seeing objects through the language glass, J. Cogn. Neurosci. 25 (10) (2013) 1702–1710

  3. [11]

    Boutonnet, G

    B. Boutonnet, G. Lupyan, Words jump-start vision: A label advantage in object recognition, J. Neurosci. 35 (25) (2015) 9329–9335

  4. [12]

    S. Hu, C. Zheng, Z. Zhou, C. Chen, G. Sukthankar, Lamp: Leveraging language prompts for multi-person pose estimation, in: Proc. IEEE/RSJ Int. Conf. Intell. Robots Syst. (IROS), IEEE, 2023, pp. 3759–3766

  5. [13]

    Zhang, W

    X. Zhang, W. Wang, Z. Chen, Y . Xu, J. Zhang, D. Tao, Clamp: Prompt-based con- trastive learning for connecting language and animal pose, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2023, pp. 23272–23281

  6. [14]

    J. Lu, D. Batra, D. Parikh, S. Lee, Vilbert: Pretraining task-agnostic visiolin- guistic representations for vision-and-language tasks, in: Advances in Neural Inf. Process. Syst. (NIPS), V ol. 32, 2019

  7. [15]

    De Bem, A

    R. De Bem, A. Arnab, S. Golodetz, M. Sapienza, P. Torr, Deep fully-connected part-based models for human pose estimation, in: Proc. Asian Conf. Mach. Learn. (ACML), PMLR, 2018, pp. 327–342. 21

  8. [16]

    W. Tang, P. Yu, Y . Wu, Deeply learned compositional models for human pose estimation, in: Proc. Eur. Conf. Comput. Vis. (ECCV), 2018, pp. 190–206

  9. [17]

    Zhang, C

    X. Zhang, C. Li, W. Hu, X. Tong, S. Maybank, Y . Zhang, Human pose estimation and tracking via parsing a tree structure based human model, IEEE Trans. Syst. Man Cybern. Syst. 44 (5) (2013) 580–592

  10. [18]

    Jun, J.-H

    J. Jun, J.-H. Lee, C.-S. Kim, Human pose estimation using skeletal heatmaps, in: Proc. Asia-Pacific Signal Inf. Process. Assoc. Annual Summit Conf. (APSIPA ASC), IEEE, 2020, pp. 1287–1292

  11. [19]

    Anderson, Q

    P. Anderson, Q. Wu, D. Teney, J. Bruce, M. Johnson, N. Sünderhauf, I. Reid, S. Gould, A. Van Den Hengel, Vision-and-language navigation: Interpret- ing visually-grounded navigation instructions in real environments, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR...

  12. [20]

    Antol, A

    S. Antol, A. Agrawal, J. Lu, M. Mitchell, D. Batra, C. L. Zitnick, D. Parikh, Vqa: Visual question answering, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2015, pp. 2425–2433

  13. [21]

    J. Li, J. Zhang, D. Tao, Referring image matting, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2023, pp. 22448–22457

  14. [22]

    T. Qiao, J. Zhang, D. Xu, D. Tao, Mirrorgan: Learning text-to-image genera- tion by redescription, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2019, pp. 1505–1514

  15. [23]

    Fürst, E

    A. Fürst, E. Rumetshofer, J. Lehner, V . T. Tran, F. Tang, H. Ramsauer, D. Kreil, M. Kopp, G. Klambauer, A. Bitto, et al., Cloob: Modern hopfield networks with infoloob outperform clip, Advances in Neural Inf. Process. Syst. (NIPS) 35 (2022) 20450–20468

  16. [24]

    C. Jia, Y . Yang, Y . Xia, Y .-T. Chen, Z. Parekh, H. Pham, Q. Le, Y .-H. Sung, Z. Li, T. Duerig, Scaling up visual and vision-language representation learning with noisy text supervision, in: Proc. Int. Conf. Mach. Learn. (ICML), PMLR, 2021, pp. 4904–4916. 22

  17. [25]

    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: Proc. Int. Conf. Mach. Learn. (ICML), PMLR, 2021, pp. 8748–8763

  18. [26]

    Zhang, M

    H. Zhang, M. C. Leong, L. Li, W. Lin, Pevl: Pose-enhanced vision-language model for fine-grained human action recognition, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2024, pp. 18857–18867

  19. [27]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, I. Polosukhin, Attention is all you need, in: Advances in Neural Inf. Process. Syst. (NIPS), 2017

  20. [28]

    Newell, K

    A. Newell, K. Yang, J. Deng, Stacked hourglass networks for human pose esti- mation, in: Proc. Eur. Conf. Comput. Vis. (ECCV), Springer, 2016, pp. 483–499

  21. [29]

    W. Yang, S. Li, W. Ouyang, H. Li, X. Wang, Learning feature pyramids for human pose estimation, in: Proc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV), 2017, pp. 1281–1290

  22. [30]

    B. Xiao, H. Wu, Y . Wei, Simple baselines for human pose estimation and tracking, in: Proc. Eur. Conf. Comput. Vis. (ECCV), 2018, pp. 466–481

  23. [31]

    K. Sun, B. Xiao, D. Liu, J. Wang, Deep high-resolution representation learn- ing for human pose estimation, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2019, pp. 5693–5703

  24. [32]

    Y . Xu, J. Zhang, Q. Zhang, D. Tao, Vitpose: simple vision transformer baselines for human pose estimation, in: Advances in Neural Inf. Process. Syst. (NIPS), 2022, pp. 38571–38584

  25. [33]

    Khirodkar, V

    R. Khirodkar, V . Chari, A. Agrawal, A. Tyagi, Multi-instance pose networks: Rethinking top-down pose estimation, in: Proc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV), 2021, pp. 3122–3131. 23

  26. [34]

    P. Lu, T. Jiang, Y . Li, X. Li, K. Chen, W. Yang, Rtmo: towards high-performance one-stage real-time multi-person pose estimation, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2024, pp. 1491–1500

  27. [35]

    Z. Geng, K. Sun, B. Xiao, Z. Zhang, J. Wang, Bottom-up human pose estimation via disentangled keypoint regression, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2021, pp. 14676–14686

  28. [36]

    Redmon, A

    J. Redmon, A. Farhadi, Yolov3: An incremental improvement, arXiv:1804.02767 (2018)

  29. [37]

    Andriluka, L

    M. Andriluka, L. Pishchulin, P. Gehler, B. Schiele, 2d human pose estimation: New benchmark and state of the art analysis, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2014, pp. 3686–3693

  30. [38]

    J. Li, C. Wang, H. Zhu, Y . Mao, H.-S. Fang, C. Lu, Crowdpose: Efficient crowded scenes pose estimation and a new benchmark, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2019, pp. 10863–10872

  31. [39]

    Zhang, R

    S.-H. Zhang, R. Li, X. Dong, P. Rosin, Z. Cai, X. Han, D. Yang, H. Huang, S.-M. Hu, Pose2seg: Detection free human instance segmentation, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2019, pp. 889–898

  32. [40]

    H. Yu, Y . Xu, J. Zhang, W. Zhao, Z. Guan, D. Tao, Ap-10k: A benchmark for animal pose estimation in the wild, arXiv preprint arXiv:2108.12617 (2021)

  33. [41]

    J. Cao, H. Tang, H.-S. Fang, X. Shen, C. Lu, Y .-W. Tai, Cross-domain adaptation for animal pose estimation, in: Proc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV), 2019, pp. 9498–9507

  34. [42]

    T.-Y . Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D. Ramanan, P. Dollár, C. L. Zitnick, Microsoft coco: Common objects in context, in: Proc. Eur. Conf. Comput. Vis. (ECCV), Springer, 2014, pp. 740–755

  35. [43]

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, L. Fei-Fei, Imagenet: A large-scale hierarchical image database, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), Ieee, 2009, pp. 248–255. 24

  36. [44]

    Loshchilov, F

    I. Loshchilov, F. Hutter, Decoupled weight decay regularization, arXiv preprint arXiv:1711.05101 (2017)

  37. [45]

    Y . Yuan, R. Fu, L. Huang, W. Lin, C. Zhang, X. Chen, J. Wang, Hrformer: high- resolution transformer for dense prediction, in: Advances in Neural Inf. Process. Syst. (NIPS), 2021, pp. 7281–7293. 25

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.