Pith. sign in

REVIEW 3 major objections 4 minor 160 references

Doodle Your Keypoints: Sketch-Based Few-Shot Keypoint Detection

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · 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 few annotated sketches can act as the only training signal for few-shot keypoint detection, letting a model localize both known and never-seen keypoints on photos of classes it has never encountered, with no photo…

desk verdict A plausible new task setup with a solid main comparison, but the real-sketch generalization claim rests on a thin evaluation that uses training photos. read the letter →

arxiv 2507.07994 v3 pith:PNNM577L submitted 2025-07-10 cs.CV

Subhajit Maity , Ayan Kumar Bhunia , Subhadeep Koley , Pinaki Nath Chowdhury , Aneeshan Sain , Yi-Zhe Song This is my paper · ORCID
classification cs.CV
keywords few-shotkeypointdetectionsketch-basedlearningcross-modalsource-freedomainadaptationprototypicalnetworkslocalizationgrid-basedlocatoredgemaps
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper claims that a few annotated sketches can replace photo data entirely in few-shot keypoint detection, giving a source-free setup. Training on base classes with edge-map supports, the model locates both base and novel keypoints on query photos, including classes never seen during training. The reported gains over a photo-designed baseline repurposed to edge maps are about 5 PCK points on Animal Pose and 4 on Animal Kingdom for novel keypoints on unseen classes. If the claim holds, a user could mark novel keypoints on a doodle of a new species and the system would find those points on real photographs.

What carries the argument

The machinery is a prototypical few-shot pipeline adapted to keypoint geometry. Gaussian pooling extracts keypoint embeddings from support edge-map features; a de-stylization network Z, built from an attention-style fusion of local keypoint context and global feature context, maps them to style-agnostic embeddings that are averaged into prototypes. Each prototype is correlated with query features by elementwise multiplication, refined by a descriptor network, and localized by the Grid Based Locator (GBL), which splits localization into grid classification with cross-entropy and within-grid offset regression with an L1 loss. Two losses carry the cross-modal burden: a supervised reduction of the prototype transport loss that matches prototypes to query keypoint embeddings at the keypoint level, and a style loss that pulls together embeddings of the same keypoint extracted from PiDiNet, HED, and Canny edge maps of the same photo.

What would settle it

Evaluate the trained model on real free-hand sketches whose query photos come from held-out novel classes and compare PCK@0.1 against the 39.0 reported with edge-map supports; the paper's own real-sketch numbers were measured on query photos from the training set. If real sketches on genuinely unseen photos fall well below the reported figures, the source-free claim for hand drawings is not established, and the edge-map-to-sketch premise would need direct evidence.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that keypoint detection admits a source-free few-shot formulation in which hand-drawn or edge-map sketches serve as the support modality. Support sketches are encoded by a ResNet50 backbone, keypoint embeddings are extracted by Gaussian pooling, averaged into per-keypoint prototypes, and correlated with query photo features by elementwise multiplication; a descriptor network followed by a grid classifier and grid offset regressor localizes each keypoint, while a supervised adaptation of the prototype transport loss and a style-disentangling loss handle the sketch-photo domain gap. The empirical claim is that this framework beats the photo-designed FSKD method repurposed to edge-map support by roughly five PCK points on Animal Pose and four on Animal Kingdom for novel keypoints on unseen classes, and that real free-hand sketches from the Sketchy database come within about half to two PCK points of edge-map supports in the settings tested.

Load-bearing premise

The central bet is that edge maps produced by off-the-shelf detectors faithfully stand in for real free-hand sketches, including the style and sparsity variety real sketchers produce, so that training on synthetic edge maps transfers to actual hand drawings — a transfer tested only on 30 real sketches whose query photos already appear in the training set.

Editorial extensions

If this is right

  • A user who sketches a novel class with keypoints marked can transfer the model to that class without any photo of it, because training needs only edge maps derived from photos of base classes.
  • The main table reports the model beating the repurposed FSKD baseline by roughly 5 PCK points on Animal Pose and 4 on Animal Kingdom for novel keypoints on unseen classes (39.0 vs 33.9 and 14.4 vs 10.1).
  • Real free-hand sketches from the Sketchy database come within roughly 0.5 to 1.8 PCK points of edge-map supports across the four evaluation settings, so hand drawings are a workable prompt interface.
  • Joint training with photos and sketches reaches 46.54 PCK for novel keypoints on unseen classes, above photo-only FSKD's 44.75, so sketches add signal rather than merely substituting for photos.

Reading between the lines

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

  • Editorial: the de-stylization loss, which forces matching embeddings across three edge detectors, reads like a general recipe for style-agnostic keypoint learning; the same trick could plausibly extend to pencil drawings, clipart, or cartoon line art without architectural changes.
  • Editorial: the paper's own text-support experiment scored 20.13 PCK, far below the sketch-based 39.0, suggesting the machinery is specific to image-like line inputs, so a genuinely source-free multimodal version would need new design rather than a drop-in encoder.
  • Editorial: the strength of the source-free claim on true novelty is untested, since the 30 real-sketch evaluations used query photos from the training set; evaluating real hand sketches on fully held-out classes would be the natural next experiment.
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 / 4 minor

Summary. The paper proposes a few-shot keypoint detection framework that uses a few annotated sketches (or their synthetic edgemap proxies) as support and localizes keypoints on query photos, including novel keypoints on unseen classes. The method builds support keypoint prototypes via Gaussian pooling, correlates them with query features through elementwise modulation, and localizes via a grid-based locator. Two additional components are introduced: a prototypical domain adaptation loss that pulls query keypoint embeddings toward support prototypes, and a de-stylization network with a style-consistency loss across multiple edgemap styles. Experiments on Animal Pose and Animal Kingdom compare against a repurposed FSKD baseline and several incremental ablations, reporting consistent gains, with the headline result being 39.0 vs. 33.9 PCK@0.1 for novel keypoints on unseen classes on Animal Pose.

Significance. If the central claim holds, the paper establishes a genuinely new capability: detecting novel keypoints on photos from sketch-only support, without photo support at inference time. The framework is well motivated, and the paper gives credit where due by carefully adapting FSKD's grid-based locator and by introducing task-appropriate modifications to prototypical domain adaptation. The strongest evidence is the consistent margin over the repurposed FSKD baseline across both datasets and across all four evaluation settings. The ablations are informative: the incremental baselines in Table 2, the de-stylization architecture study in Supplementary Table 3, and the edge-detector combination study in Supplementary Table 4 isolate the contribution of each component. The paper is less convincing on the load-bearing premise that edgemaps are an adequate proxy for real free-hand sketches, because the real-sketch evaluation is small, possibly overlapping with training images, and supplemented only by subjective ratings.

major comments (3)
  1. [Sec. 4.2 and Suppl. Sec. 9 (Table 5)] The real free-hand sketch evaluation does not establish held-out cross-modal generalization. The main text says the evaluation is performed on 'query photos in the training dataset [15]' using 30 real sketches, and the supplementary repeats this without clarifying whether those query photos were excluded from episodic training. If the query photos overlap the training split, the reported PCK values (38.49 for unseen-class novel, 44.99 for seen-class novel) can be inflated by image-level memorization, so they do not measure transfer to real sketches on unseen photos. Please re-run the evaluation on held-out photos, report the exact split, and provide per-class sample counts and variance across random support draws.
  2. [Sec. 3, first paragraph] The framework's central premise is stated as 'we use off-the-shelf edge detectors ... considering them as equivalent to sketch data.' This equivalence is load-bearing because the de-stylization loss and domain adaptation are trained only on detector-generated styles. The current quantitative bridge to real sketches is the small 30-sketch evaluation discussed above, plus the subjective human study in Suppl. Sec. 11, which records 1-5 ratings rather than localization accuracy against ground truth. As written, the paper does not provide a clean quantitative test that the edgemap-to-hand-drawing proxy transfers to user-specific sparsity and abstraction; I would like to see a held-out real-sketch evaluation or an explicit acknowledgment that real-sketch generalization remains an open validation.
  3. [Sec. 5 and Abstract] The phrase 'source-free setup' is stronger than what is demonstrated. The method uses supervised episodic training on base classes with annotated query photos, and Eq. (10) uses query keypoint embeddings extracted from annotated query photos; only at inference is the support set sketch-only. This is a meaningful contribution, but the paper should state precisely what 'source-free' means here and avoid implying that no annotated source domain was used.
minor comments (4)
  1. [Sec. 4, Implementation Details] The paper reports a single run for each configuration; because the episodic few-shot protocol is stochastic, please report mean and standard deviation over multiple random seeds or episodes.
  2. [Sec. 4.2] Thirty real sketches across five classes is only about six sketches per class; the small sample size should be stated as a limitation, and it would be useful to know how many distinct Sketchy drawings and how many participants were used.
  3. [Suppl. Sec. 11] The human study measures perceived quality, not localization accuracy, so the sentence 'this study further proves the practicality of our framework' overstates what the ratings establish.
  4. [Sec. 3.2, Eq. (10)] The derivation from the bidirectional transport loss of Tanwisuth et al. to Eq. (10) is compressed; the supplementary reconstruction is helpful, but the main text should clarify that p(hat_mu_n) is computed from query annotations during training and is therefore a supervised, not unsupervised, target distribution.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: predictions are computed from support-only prototypes, and the reported gains are measured against repurposed baselines rather than derived from fitted parameters.

full rationale

Walking the claimed derivation chain: support prototypes are computed by Eq. (3) as means of de-stylized keypoint embeddings extracted only from support sketches or edgemaps; query labels enter only through training losses and never through the inference-time prototypes. The localization head is explicitly adapted from Lu et al. [67] with cross-entropy and L1 regression objectives (Eqs. 6-8), which is borrowing from external prior work, not circular. The domain-adaptation loss is reduced step-by-step from the bi-directional transport loss of Tanwisuth et al. [115] in Eqs. (12)-(14) of the supplement, with the paper stating which weighting and softmax terms are dropped; this is a stated adaptation of an external method, not a self-citation chain. The de-stylization network is validated by ablations (Tables 2 and 3), and the premise that off-the-shelf edgemaps proxy sketches is an explicit assumption ('we use off-the-shelf edge detectors... considering them as equivalent to sketch data') that is testable and is in fact tested against real Sketchy sketches. Whether that real-sketch test is fully clean, since the main text says evaluation uses 'query photos in the training dataset [15]', is a generalization and experimental-design concern, not circularity: the model is not fitted to those real sketches and no equation in the paper reduces to a fitted parameter renamed as a prediction. The many SketchX self-citations appear in related-work positioning and are not load-bearing; the load-bearing references (FSKD [67], prototypical UDA [115], Prototypical Networks [108], PiDiNet/HED/Canny) are external and independently checkable. The central claim is therefore self-contained with respect to circularity.

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

The framework rests on a small set of training-proxy assumptions: edgemaps stand in for sketches, ImageNet-pretrained features transfer across the sketch-photo gap, prototype transport aligns keypoint embeddings, auxiliary interpolation is useful, and base-class knowledge transfers to novel keypoints. No new physical entities are postulated; the de-stylization network and losses are functions over existing feature maps. The main untested premise is the equivalence between generated edgemaps and real free-hand sketches, which is mitigated only by a 30-sketch evaluation on training-distribution photos.

free parameters (5)
  • Gaussian pooling scale xi = 1/4
    Controls the soft local context around keypoints in Eq. (2); chosen empirically in Sec. 4.
  • Loss weights lambda_KP, lambda_DA, lambda_style = 0.5, 0.001, 0.001
    Balance localization, domain adaptation, and de-stylization in Eq. (1); empirically determined in Sec. 4.
  • GBL grid scales L = {8, 12, 16}
    Multi-scale grid levels for the grid classifier and regressor; empirically determined in Sec. 4.
  • Edge detector set for synthetic sketches = {PiDiNet, HED, Canny}
    Selected after experimenting with many combinations and choosing the best on the same evaluation setting (Suppl. Sec. 8, Tab. 4), so the reported headline number is partially tuned.
  • Auxiliary keypoint configuration = 6 pairs, t in {0.25, 0.5, 0.75}
    Interpolation points and pairs for auxiliary keypoints; design choice that provides the largest single performance jump in Tab. 2.
assumptions (5)
  • ad hoc to paper Edgemaps from PiDiNet, HED, and Canny are an adequate stand-in for free-hand sketch styles.
    Training uses only these synthetic sketches; the real-sketch validation is a small, non-held-out test, so this equivalence is load-bearing and not independently established.
  • domain assumption A single ImageNet-pretrained ResNet50 encoder produces features that transfer across sketch and photo domains.
    No cross-modal pretraining or domain-specific alignment is shown before the prototype and domain-adaptation stages.
  • domain assumption Prototype-based transport loss aligns keypoint embeddings across the sketch-photo gap.
    Borrowed from [115] and adapted from classification to keypoints; the adaptation is justified by task-level reasoning, not by a derivation or an external validation.
  • domain assumption Interpolated auxiliary keypoints, with visibility provided by a saliency detector, are valid keypoint supervision.
    Auxiliary keypoints provide the largest single performance jump in Tab. 2, so their validity is important; the saliency detector is taken off-the-shelf.
  • domain assumption Base keypoint knowledge transfers to novel keypoints and unseen classes in the episodic protocol.
    This is the standard few-shot transfer assumption and is tested empirically, not proven.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Doodle Your Keypoints: Sketch-Based Few-Shot Keypoint Detection." pith.science (2026). https://pith.science/paper/PNNM577L

@misc{pith2026250707994,
  author       = {Pith},
  title        = {Pith review of: Doodle Your Keypoints: Sketch-Based Few-Shot Keypoint Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PNNM577L}},
  note         = {Machine review of arXiv:2507.07994}
}
read the original abstract

Keypoint detection, integral to modern machine perception, faces challenges in few-shot learning, particularly when source data from the same distribution as the query is unavailable. This gap is addressed by leveraging sketches, a popular form of human expression, providing a source-free alternative. However, challenges arise in mastering cross-modal embeddings and handling user-specific sketch styles. Our proposed framework overcomes these hurdles with a prototypical setup, combined with a grid-based locator and prototypical domain adaptation. We also demonstrate success in few-shot convergence across novel keypoints and classes through extensive experiments.

Figures

Figures reproduced from arXiv: 2507.07994 by the authors.

Figure 1
Figure 1. Keypoint detection [17], usually approached in a supervised setting [29], along with cross-domain adaptation [15], can be set up in a few-shot paradigm [123] by typically learning to localize novel keypoints from limited annotated photos [41], also adapting to novel classes [67]. Unlike the existing works, we approach the problem in a cross-modal setup. (right) The proposed few-shot framework adapts to localize nove… view at source ↗
Figure 2
Figure 2. Overview of the proposed few-shot key-point detection framework that processes sketches or edgemaps in the support set and [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. The design of the de-stylization network [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Sample support (S) and query (Q) for training (top) and evaluation (bottom) for all evaluation settings. Evaluation Protocol: We use percentage correct key￾points (PCK) [81] with a margin τ = 0.1 as metric. The prediction is correct if within a range τ times the larger…
Figure 5
Figure 5. Figure 5: Visualising detection (✖) and ground-truth (●) for novel keypoints for base classes (top) and unseen classes (bottom). 4.1. Performance Analysis A detailed performance comparison of the naive baseline B-Vanilla and FSKD [67] (repurposed for the cross-modal paradigm) wi…
Figure 6
Figure 6. Figure 6: Inference (✖) with ground-truth (●) for base (top) and novel (bottom) keypoints with annotated sketch prompts. ground truth from Animal Pose [15] are presented in [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Performance with varying shots. 3 5 7 9 11 20 30 40 N PCK@0.1 [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 10
Figure 10. Figure 10: Visualizations of sample detection (✖) along with ground-truth (●) for base (left) and novel (right) keypoints on query pho￾tos [15] using support edgemaps [110]. ing on the de-stylization loss Lstyle (with reduced weight of λstyle = 10−8 ) to learn the keypoint repre…
Figure 11
Figure 11. Figure 11: Visualizations of sample detection (✖) along with ground-truth (●) for base (left) and novel (right) keypoints on query pho￾tos [15] using real support sketches [104] with manual annotation prompt. tual encoder [89] to obtain textual keypoint embeddings that replace s…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

160 extracted references · 74 canonical work pages

  1. [15]

    Cross-domain adapta- tion for animal pose estimation

    Jinkun Cao, Hongyang Tang, Hao-Shu Fang, Xiaoyong Shen, Cewu Lu, and Yu-Wing Tai. Cross-domain adapta- tion for animal pose estimation. In ICCV, 2019. 1, 2, 3, 6, 7, 15, 16, 17, 18

    2019

  2. [1]

    Deep learning using rectified linear units (relu)

    Abien Fred Agarap. Deep learning using rectified linear units (relu). arXiv preprint arXiv:1803.08375, 2018. 6

    arXiv 2018

  3. [2]

    Allen, Evan Shelhamer, Hanul Shin, and Joshua B

    Kelsey R. Allen, Evan Shelhamer, Hanul Shin, and Joshua B. Tenenbaum. Infinite mixture prototypes for few- shot learning. In ICML, 2019. 2

    2019

  4. [3]

    Coke: Contrastive learning for robust keypoint de- tection

    Yutong Bai, Angtian Wang, Adam Kortylewski, and Alan Yuille. Coke: Contrastive learning for robust keypoint de- tection. In WACV, 2023. 3

    2023

  5. [4]

    Surf: Speeded up robust features

    Herbert Bay, Tinne Tuytelaars, and Luc Van Gool. Surf: Speeded up robust features. In ECCV, 2006. 1, 2

    2006

  6. [5]

    Towards Practicality of Sketch-Based Visual Understanding

    Ayan Kumar Bhunia. Towards Practicality of Sketch-Based Visual Understanding. PhD thesis, University of Surrey,

  7. [6]

    Pixelor: A competitive sketching ai agent

    Ayan Kumar Bhunia, Ayan Das, Umar Riaz Muhammad, Yongxin Yang, Timothy M Hospedales, Tao Xiang, Yulia Gryaditskaya, and Yi-Zhe Song. Pixelor: A competitive sketching ai agent. so you think you can sketch? Siggraph Asia, 2020. 2

    2020

  8. [7]

    Sketch less for more: On-the-fly fine-grained sketch-based image retrieval

    Ayan Kumar Bhunia, Yongxin Yang, Timothy M Hospedales, Tao Xiang, and Yi-Zhe Song. Sketch less for more: On-the-fly fine-grained sketch-based image retrieval. In CVPR, 2020. 2, 3, 17

    2020

Show all 160 references
  1. [8]

    More photos are all you need: Semi-supervised learning for fine- grained sketch based image retrieval

    Ayan Kumar Bhunia, Pinaki Nath Chowdhury, Aneeshan Sain, Yongxin Yang, Tao Xiang, and Yi-Zhe Song. More photos are all you need: Semi-supervised learning for fine- grained sketch based image retrieval. In CVPR, 2021. 2, 3, 17

    2021

  2. [9]

    Doodle it yourself: Class incremental learning by drawing a few sketches

    Ayan Kumar Bhunia, Viswanatha Reddy Gajjala, Sub- hadeep Koley, Rohit Kundu, Aneeshan Sain, Tao Xiang, and Yi-Zhe Song. Doodle it yourself: Class incremental learning by drawing a few sketches. In CVPR, 2022. 2, 3

    2022

  3. [10]

    Sketch2saliency: Learning to detect salient objects from human drawings

    Ayan Kumar Bhunia, Subhadeep Koley, Amandeep Kumar, Aneeshan Sain, Pinaki Nath Chowdhury, Tao Xiang, and Yi-Zhe Song. Sketch2saliency: Learning to detect salient objects from human drawings. In CVPR, 2023. 2, 3

    2023

  4. [11]

    3fabrec: Fast few-shot face alignment by reconstruction

    Bjorn Browatzki and Christian Wallraven. 3fabrec: Fast few-shot face alignment by reconstruction. In CVPR, 2020. 2, 3

    2020

  5. [12]

    Memory matching networks for one-shot image recognition

    Qi Cai, Yingwei Pan, Ting Yao, Chenggang Yan, and Tao Mei. Memory matching networks for one-shot image recognition. In CVPR, 2018. 2

    2018

  6. [13]

    Cascade r-cnn: Delv- ing into high quality object detection

    Zhaowei Cai and Nuno Vasconcelos. Cascade r-cnn: Delv- ing into high quality object detection. In CVPR, 2018. 2

    2018

  7. [14]

    A computational approach to edge detection

    John Canny. A computational approach to edge detection. IEEE TPAMI, 1986. 3, 5, 8, 16

    1986

  8. [16]

    Realtime multi-person 2d pose estimation using part affin- ity fields

    Zhe Cao, Tomas Simon, Shih-En Wei, and Yaser Sheikh. Realtime multi-person 2d pose estimation using part affin- ity fields. In CVPR, 2017. 2

    2017

  9. [17]

    Openpose: realtime multi-person 2d pose estimation using part affinity fields

    Zhe Cao, Gines Hidalgo, Tomas Simon, Shih-En Wei, and Yaser Sheikh. Openpose: realtime multi-person 2d pose estimation using part affinity fields. IEEE TPAMI, 2021. 1, 2, 3

    2021

  10. [18]

    Human pose estimation with iterative error feedback

    Joao Carreira, Pulkit Agrawal, Katerina Fragkiadaki, and Jitendra Malik. Human pose estimation with iterative error feedback. In CVPR, 2016. 1, 2

    2016

  11. [19]

    Higherhrnet: Scale- aware representation learning for bottom-up human pose estimation

    Bowen Cheng, Bin Xiao, Jingdong Wang, Honghui Shi, Thomas S Huang, and Lei Zhang. Higherhrnet: Scale- aware representation learning for bottom-up human pose estimation. In CVPR, 2020. 1, 3

    2020

  12. [20]

    Partially does it: Towards scene-level fg-sbir with partial input

    Pinaki Nath Chowdhury, Ayan Kumar Bhunia, Viswanatha Reddy Gajjala, Aneeshan Sain, Tao Xiang, and Yi-Zhe Song. Partially does it: Towards scene-level fg-sbir with partial input. In CVPR, 2022. 2, 17

    2022

  13. [21]

    What can human sketches do for object detection? In CVPR,

    Pinaki Nath Chowdhury, Ayan Kumar Bhunia, Aneeshan Sain, Subhadeep Koley, Tao Xiang, and Yi-Zhe Song. What can human sketches do for object detection? In CVPR,

  14. [22]

    Democratising 2d sketch to 3d shape retrieval through piv- oting

    Pinaki Nath Chowdhury, Ayan Kumar Bhunia, Aneeshan Sain, Subhadeep Koley, Tao Xiang, and Yi-Zhe Song. Democratising 2d sketch to 3d shape retrieval through piv- oting. In ICCV, 2023. 2

    2023

  15. [23]

    Scenetrilogy: On human scene-sketch and its complemen- tarity with photo and text

    Pinaki Nath Chowdhury, Ayan Kumar Bhunia, Aneeshan Sain, Subhadeep Koley, Tao Xiang, and Yi-Zhe Song. Scenetrilogy: On human scene-sketch and its complemen- tarity with photo and text. In CVPR, 2023. 2

    2023

  16. [24]

    Livesketch: Query perturbations for guided sketch-based visual search

    John Collomosse, Tu Bui, and Hailin Jin. Livesketch: Query perturbations for guided sketch-based visual search. In CVPR, 2019. 2

    2019

  17. [25]

    Attentional feature fusion

    Yimian Dai, Fabian Gieseke, Stefan Oehmcke, Yiquan Wu, and Kobus Barnard. Attentional feature fusion. In WACV,

  18. [26]

    Imagenet: A large-scale hierarchical im- age database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical im- age database. In CVPR, 2009. 6

    2009

  19. [27]

    Teacher supervises students how to learn from partially labeled images for facial land- mark detection

    Xuanyi Dong and Yi Yang. Teacher supervises students how to learn from partially labeled images for facial land- mark detection. In ICCV, 2019. 3

    2019

  20. [28]

    Semantically tied paired cycle consistency for zero-shot sketch-based image re- trieval

    Anjan Dutta and Zeynep Akata. Semantically tied paired cycle consistency for zero-shot sketch-based image re- trieval. In CVPR, 2019. 2

    2019

  21. [29]

    Rmpe: Regional multi-person pose estimation

    Hao-Shu Fang, Shuqin Xie, Yu-Wing Tai, and Cewu Lu. Rmpe: Regional multi-person pose estimation. In ICCV,

  22. [30]

    Model- agnostic meta-learning for fast adaptation of deep networks

    Chelsea Finn, Pieter Abbeel, and Sergey Levine. Model- agnostic meta-learning for fast adaptation of deep networks. In ICML, 2017. 2

    2017

  23. [31]

    Low-shot learning via covariance- preserving adversarial augmentation networks

    Hang Gao, Zheng Shou, Alireza Zareian, Hanwang Zhang, and Shih-Fu Chang. Low-shot learning via covariance- preserving adversarial augmentation networks. In NeurIPS,

  24. [32]

    Creative sketch generation

    Songwei Ge, Vedanuj Goswami, Larry Zitnick, and Devi Parikh. Creative sketch generation. In ICLR, 2020. 2

    2020

  25. [33]

    Metacloth: Learning unseen tasks of dense fashion landmark detection from a few samples

    Yuying Ge, Ruimao Zhang, and Ping Luo. Metacloth: Learning unseen tasks of dense fashion landmark detection from a few samples. IEEE TIP, 2021. 3, 18

    2021

  26. [34]

    Interactive sketch & fill: Multiclass sketch-to-image trans- lation

    Arnab Ghosh, Richard Zhang, Puneet K Dokania, Oliver Wang, Alexei A Efros, Philip HS Torr, and Eli Shechtman. Interactive sketch & fill: Multiclass sketch-to-image trans- lation. In ICCV, 2019. 2 9

    2019

  27. [35]

    Neural tur- ing machines

    Alex Graves, Greg Wayne, and Ivo Danihelka. Neural tur- ing machines. arXiv preprint arXiv:1410.5401, 2014. 2

    2014 arXiv

  28. [36]

    Keypoint transformer: Solving joint identifi- cation in challenging hands and object interactions for ac- curate 3d pose estimation

    Shreyas Hampali, Sayan Deb Sarkar, Mahdi Rad, and Vin- cent Lepetit. Keypoint transformer: Solving joint identifi- cation in challenging hands and object interactions for ac- curate 3d pose estimation. In CVPR, 2022. 1, 2

    2022

  29. [37]

    Low-shot visual recognition by shrinking and hallucinating features

    Bharath Hariharan and Ross Girshick. Low-shot visual recognition by shrinking and hallucinating features. In ICCV, 2017. 2

    2017

  30. [38]

    A combined corner and edge detector

    Chris Harris, Mike Stephens, et al. A combined corner and edge detector. In Alvey vision conference, 1988. 1, 2

    1988

  31. [39]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In CVPR,

  32. [40]

    Mask r-cnn

    Kaiming He, Georgia Gkioxari, Piotr Doll ´ar, and Ross Gir- shick. Mask r-cnn. In ICCV, 2017. 2

    2017

  33. [41]

    Few-shot geometry-aware keypoint lo- calization

    Xingzhe He, Gaurav Bharaj, David Ferman, Helge Rhodin, and Pablo Garrido. Few-shot geometry-aware keypoint lo- calization. In CVPR, 2023. 1, 3, 18

    2023

  34. [42]

    Why do line drawings work? a realism hypothesis

    Aaron Hertzmann. Why do line drawings work? a realism hypothesis. Perception, 2020. 1, 2, 3, 8

    2020

  35. [43]

    Improving landmark localization with semi-supervised learning

    Sina Honari, Pavlo Molchanov, Stephen Tyree, Pascal Vin- cent, Christopher Pal, and Jan Kautz. Improving landmark localization with semi-supervised learning. InCVPR, 2018. 3

    2018

  36. [44]

    Sketch-a-segmenter: Sketch-based photo segmenter generation.IEEE TIP, 2020

    Conghui Hu, Da Li, Yongxin Yang, Timothy M Hospedales, and Yi-Zhe Song. Sketch-a-segmenter: Sketch-based photo segmenter generation.IEEE TIP, 2020. 2

    2020

  37. [45]

    Acmm: Aligned cross-modal memory for few-shot image and sentence matching

    Yan Huang and Liang Wang. Acmm: Aligned cross-modal memory for few-shot image and sentence matching. In ICCV, 2019. 3

    2019

  38. [46]

    Unsupervised learning of object landmarks through conditional image generation

    Tomas Jakab, Ankush Gupta, Hakan Bilen, and An- drea Vedaldi. Unsupervised learning of object landmarks through conditional image generation. In NeurIPS, 2018. 3

    2018

  39. [47]

    Self-supervised learning of interpretable keypoints from unlabelled videos

    Tomas Jakab, Ankush Gupta, Hakan Bilen, and Andrea Vedaldi. Self-supervised learning of interpretable keypoints from unlabelled videos. In CVPR, 2020. 3

    2020

  40. [48]

    Adam: a method for stochastic optimization

    DP Kingma. Adam: a method for stochastic optimization. In ICLR, 2014. 6

    2014

  41. [49]

    Siamese neural networks for one-shot image recognition

    Gregory Koch, Richard Zemel, and Ruslan Salakhutdinov. Siamese neural networks for one-shot image recognition. In ICML Deep Learning Workshop, 2015. 2

    2015

  42. [50]

    Pic- ture that sketch: Photorealistic image generation from ab- stract sketches

    Subhadeep Koley, Ayan Kumar Bhunia, Aneeshan Sain, Pinaki Nath Chowdhury, Tao Xiang, and Yi-Zhe Song. Pic- ture that sketch: Photorealistic image generation from ab- stract sketches. In CVPR, 2023. 2

    2023

  43. [51]

    Human-level concept learning through proba- bilistic program induction

    Brenden M Lake, Ruslan Salakhutdinov, and Joshua B Tenenbaum. Human-level concept learning through proba- bilistic program induction. Science, 2015. 2

    2015

  44. [52]

    Cornernet: Detecting objects as paired keypoints

    Hei Law and Jia Deng. Cornernet: Detecting objects as paired keypoints. In ECCV, 2018. 2

    2018

  45. [53]

    Few-shot learning with global class representa- tions

    Aoxue Li, Tiange Luo, Tao Xiang, Weiran Huang, and Li- wei Wang. Few-shot learning with global class representa- tions. In ICCV, 2019. 2

    2019

  46. [54]

    Cross-modal learning with adversarial samples

    Chao Li, Shangqian Gao, Cheng Deng, De Xie, and Wei Liu. Cross-modal learning with adversarial samples. In NeurIPS, 2019. 3

    2019

  47. [55]

    Photo pre-training, but for sketch

    Ke Li, Kaiyue Pang, and Yi-Zhe Song. Photo pre-training, but for sketch. In CVPR, 2023. 17

    2023

  48. [56]

    Photo-sketching: Inferring contour draw- ings from images

    Mengtian Li, Zhe Lin, Radomir Mech, Ersin Yumer, and Deva Ramanan. Photo-sketching: Inferring contour draw- ings from images. In WACV, 2019. 2, 16

    2019

  49. [57]

    Deformation-aware un- paired image translation for pose estimation on laboratory animals

    Siyuan Li, Semih Gunel, Mirela Ostrek, Pavan Ramdya, Pascal Fua, and Helge Rhodin. Deformation-aware un- paired image translation for pose estimation on laboratory animals. In CVPR, 2020. 3

    2020

  50. [58]

    Revisiting local descriptor based image-to-class measure for few-shot learning

    Wenbin Li, Lei Wang, Jinglin Xu, Jing Huo, Yang Gao, and Jiebo Luo. Revisiting local descriptor based image-to-class measure for few-shot learning. In CVPR, 2019. 2

    2019

  51. [59]

    Deep sketch-guided cartoon video inbetweening

    Xiaoyu Li, Bo Zhang, Jing Liao, and Pedro V Sander. Deep sketch-guided cartoon video inbetweening. IEEE TVCG,

  52. [60]

    Im2pencil: Controllable pencil il- lustration from photographs

    Yijun Li, Chen Fang, Aaron Hertzmann, Eli Shechtman, and Ming-Hsuan Yang. Im2pencil: Controllable pencil il- lustration from photographs. In CVPR, 2019. 2, 16

    2019

  53. [61]

    Tokenpose: Learning keypoint tokens for human pose estimation

    Yanjie Li, Shoukui Zhang, Zhicheng Wang, Sen Yang, Wankou Yang, Shu-Tao Xia, and Erjin Zhou. Tokenpose: Learning keypoint tokens for human pose estimation. In ICCV, 2021. 1, 2

    2021

  54. [62]

    Dense classification and implanting for few-shot learning

    Yann Lifchitz, Yannis Avrithis, Sylvaine Picard, and Andrei Bursuc. Dense classification and implanting for few-shot learning. In CVPR, 2019. 2

    2019

  55. [63]

    Multimodality helps unimodality: Cross-modal few-shot learning with multimodal models

    Zhiqiu Lin, Samuel Yu, Zhiyi Kuang, Deepak Pathak, and Deva Ramanan. Multimodality helps unimodality: Cross-modal few-shot learning with multimodal models. In CVPR, 2023. 3

    2023

  56. [64]

    Disentangling and unifying graph convolutions for skeleton-based action recognition

    Ziyu Liu, Hongwen Zhang, Zhenghao Chen, Zhiyong Wang, and Wanli Ouyang. Disentangling and unifying graph convolutions for skeleton-based action recognition. In CVPR, 2020. 1

    2020

  57. [65]

    Unsupervised part-based disentangling of object shape and appearance

    Dominik Lorenz, Leonard Bereska, Timo Milbich, and Bjorn Ommer. Unsupervised part-based disentangling of object shape and appearance. In CVPR, 2019. 3

    2019

  58. [66]

    Object recognition from local scale- invariant features

    David G Lowe. Object recognition from local scale- invariant features. In ICCV, 1999. 1, 2

    1999

  59. [67]

    Few-shot keypoint de- tection with uncertainty learning for unseen species

    Changsheng Lu and Piotr Koniusz. Few-shot keypoint de- tection with uncertainty learning for unseen species. In CVPR, 2022. 1, 2, 3, 4, 5, 6, 7, 8, 14, 15, 16, 17, 18

    2022

  60. [68]

    Detect any keypoints: An efficient light-weight few-shot keypoint detector

    Changsheng Lu and Piotr Koniusz. Detect any keypoints: An efficient light-weight few-shot keypoint detector. In AAAI, 2024. 3

    2024

  61. [69]

    From saliency to dino: Saliency-guided vision transformer for few-shot keypoint detection

    Changsheng Lu, Hao Zhu, and Piotr Koniusz. From saliency to dino: Saliency-guided vision transformer for few-shot keypoint detection. arXiv preprint arXiv:2304.03140, 2023. 3, 18

    2023 arXiv

  62. [70]

    Openkd: Opening prompt diversity for zero-and few-shot keypoint detection

    Changsheng Lu, Zheyuan Liu, and Piotr Koniusz. Openkd: Opening prompt diversity for zero-and few-shot keypoint detection. In ECCV, 2024. 3, 17, 18

    2024

  63. [71]

    Learning dense correspondences between photos and sketches

    Xuanchen Lu, Xiaolong Wang, and Judith E Fan. Learning dense correspondences between photos and sketches. In ICML, 2023. 17

    2023

  64. [72]

    Rethinking the heatmap regression for bottom-up human pose estimation

    Zhengxiong Luo, Zhicheng Wang, Yan Huang, Liang Wang, Tieniu Tan, and Erjin Zhou. Rethinking the heatmap regression for bottom-up human pose estimation. In CVPR,

  65. [73]

    Deeplabcut: markerless pose estimation of user-defined body parts with deep learning

    Alexander Mathis, Pranav Mamidanna, Kevin M Cury, Taiga Abe, Venkatesh N Murthy, Mackenzie Weygandt Mathis, and Matthias Bethge. Deeplabcut: markerless pose estimation of user-defined body parts with deep learning. Nature Neuroscience, 2018. 3

    2018

  66. [74]

    A simple neural attentive meta-learner

    Nikhil Mishra, Mostafa Rohaninejad, Xi Chen, and Pieter Abbeel. A simple neural attentive meta-learner. In ICLR,

  67. [75]

    Semi-supervised keypoint localization

    Olga Moskvyak, Frederic Maire, Feras Dayoub, and Mahsa Baktashmotlagh. Semi-supervised keypoint localization. In ICLR, 2020. 3

    2020

  68. [76]

    Meta networks

    Tsendsuren Munkhdalai and Hong Yu. Meta networks. In ICML, 2017. 2

    2017

  69. [77]

    G-cnn: an iterative grid based object detector

    Mahyar Najibi, Mohammad Rastegari, and Larry S Davis. G-cnn: an iterative grid based object detector. In CVPR,

  70. [78]

    Stacked hourglass networks for human pose estimation

    Alejandro Newell, Kaiyu Yang, and Jia Deng. Stacked hourglass networks for human pose estimation. In ECCV,

  71. [79]

    Animal kingdom: A large and di- verse dataset for animal behavior understanding

    Xun Long Ng, Kian Eng Ong, Qichen Zheng, Yun Ni, Si Yong Yeo, and Jun Liu. Animal kingdom: A large and di- verse dataset for animal behavior understanding. In CVPR,

  72. [80]

    On first-order meta-learning algorithms

    Alex Nichol, Joshua Achiam, and John Schulman. On first-order meta-learning algorithms. arXiv preprint arXiv:1803.02999, 2018. 2

    2018 arXiv

  73. [81]

    Self-supervised learning of geometrically sta- ble features through probabilistic introspection

    David Novotny, Samuel Albanie, Diane Larlus, and An- drea Vedaldi. Self-supervised learning of geometrically sta- ble features through probabilistic introspection. In CVPR,

  74. [82]

    Tadam: Task dependent adaptive metric for improved few-shot learning

    Boris Oreshkin, Pau Rodr ´ıguez L ´opez, and Alexandre La- coste. Tadam: Task dependent adaptive metric for improved few-shot learning. In NeurIPS, 2018. 2

    2018

  75. [83]

    Multimodal prototypical networks for few-shot learn- ing

    Frederik Pahde, Mihai Puscas, Tassilo Klein, and Moin Nabi. Multimodal prototypical networks for few-shot learn- ing. In WACV, 2021. 3

    2021

  76. [84]

    Pytorch: An imperative style, high-performance deep learning library

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. Pytorch: An imperative style, high-performance deep learning library. In NeurIPS, 2019. 6

    2019

  77. [85]

    Few-shot image recognition with knowledge transfer

    Zhimao Peng, Zechao Li, Junge Zhang, Yan Li, Guo-Jun Qi, and Jinhui Tang. Few-shot image recognition with knowledge transfer. In ICCV, 2019. 2

    2019

  78. [86]

    Low-shot learning with imprinted weights

    Hang Qi, Matthew Brown, and David G Lowe. Low-shot learning with imprinted weights. In CVPR, 2018. 2

    2018

  79. [87]

    Aggregation via separation: Boosting facial land- mark detector with semi-supervised style translation

    Shengju Qian, Keqiang Sun, Wayne Wu, Chen Qian, and Jiaya Jia. Aggregation via separation: Boosting facial land- mark detector with semi-supervised style translation. In ICCV, 2019. 3

    2019

  80. [88]

    Siyuan Qiao, Chenxi Liu, Wei Shen, and Alan L. Yuille. Few-shot image recognition by predicting parameters from activations. In CVPR, 2018. 2

    2018

  81. [89]

    Learn- ing transferable visual models from natural language super- vision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learn- ing transferable visual models from natural language super- vision. In ICML, 2021. 18

    2021

  82. [90]

    Adaptive posterior learning: few-shot learning with a surprise-based memory module

    Tiago Ramalho and Marta Garnelo. Adaptive posterior learning: few-shot learning with a surprise-based memory module. In ICLR, 2019. 2

    2019

  83. [91]

    Optimization as a model for few-shot learning

    Sachin Ravi and Hugo Larochelle. Optimization as a model for few-shot learning. In ICLR, 2017. 2

    2017

  84. [92]

    Yolo9000: Better, faster, stronger

    Joseph Redmon and Ali Farhadi. Yolo9000: Better, faster, stronger. In CVPR, 2017. 4

    2017

  85. [93]

    Yolov3: An incremental improvement

    Joseph Redmon and Ali Farhadi. Yolov3: An incremental improvement. arXiv preprint arXiv:1804.02767, 2018

    2018 arXiv

  86. [94]

    You only look once: Unified, real-time object de- tection

    Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi. You only look once: Unified, real-time object de- tection. In CVPR, 2016. 4

    2016

  87. [95]

    Faster r-cnn: Towards real-time object detection with re- gion proposal networks

    Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real-time object detection with re- gion proposal networks. In NeurIPS, 2015. 2

    2015

  88. [96]

    Sketchformer: Transformer-based represen- tation for sketched structure

    Leo Sampaio Ferraz Ribeiro, Tu Bui, John Collomosse, and Moacir Ponti. Sketchformer: Transformer-based represen- tation for sketched structure. In CVPR, 2020. 2

    2020

  89. [97]

    Encoding in style: A stylegan encoder for image-to-image translation

    Elad Richardson, Yuval Alaluf, Or Patashnik, Yotam Nitzan, Yaniv Azar, Stav Shapiro, and Daniel Cohen-Or. Encoding in style: A stylegan encoder for image-to-image translation. In CVPR, 2021. 17

    2021

  90. [98]

    Classifier grids for robust adaptive object detec- tion

    Peter M Roth, Sabine Sternig, Helmut Grabner, and Horst Bischof. Classifier grids for robust adaptive object detec- tion. In CVPR, 2009. 4

    2009

  91. [99]

    Rusu, Dushyant Rao, Jakub Sygnowski, Oriol Vinyals, Razvan Pascanu, Simon Osindero, and Raia Had- sell

    Andrei A. Rusu, Dushyant Rao, Jakub Sygnowski, Oriol Vinyals, Razvan Pascanu, Simon Osindero, and Raia Had- sell. Meta-learning with latent embedding optimization. In ICLR, 2019. 2

    2019

  92. [100]

    Stylemeup: Towards style- agnostic sketch-based image retrieval

    Aneeshan Sain, Ayan Kumar Bhunia, Yongxin Yang, Tao Xiang, and Yi-Zhe Song. Stylemeup: Towards style- agnostic sketch-based image retrieval. In CVPR, 2021. 2, 3, 5, 17

    2021

  93. [101]

    Sketch3t: Test-time training for zero-shot sbir

    Aneeshan Sain, Ayan Kumar Bhunia, Vaishnav Potlapalli, Pinaki Nath Chowdhury, Tao Xiang, and Yi-Zhe Song. Sketch3t: Test-time training for zero-shot sbir. In CVPR,

  94. [102]

    Clip for all things zero-shot sketch-based image retrieval, fine- grained or not

    Aneeshan Sain, Ayan Kumar Bhunia, Pinaki Nath Chowd- hury, Subhadeep Koley, Tao Xiang, and Yi-Zhe Song. Clip for all things zero-shot sketch-based image retrieval, fine- grained or not. In CVPR, 2023. 2, 17

    2023

  95. [103]

    Exploiting unlabelled photos for stronger fine-grained sbir

    Aneeshan Sain, Ayan Kumar Bhunia, Subhadeep Koley, Pinaki Nath Chowdhury, Soumitri Chattopadhyay, Tao Xi- ang, and Yi-Zhe Song. Exploiting unlabelled photos for stronger fine-grained sbir. In CVPR, 2023. 2, 17

    2023

  96. [104]

    The sketchy database: learning to retrieve badly drawn bunnies

    Patsorn Sangkloy, Nathan Burnell, Cusuh Ham, and James Hays. The sketchy database: learning to retrieve badly drawn bunnies. ACM TOG, 2016. 3, 7, 16, 17, 18

    2016

  97. [105]

    Meta-learning with memory-augmented neural networks

    Adam Santoro, Sergey Bartunov, Matthew Botvinick, Daan Wierstra, and Timothy Lillicrap. Meta-learning with memory-augmented neural networks. In ICML, 2016. 2

    2016

  98. [106]

    Bronstein

    Eli Schwartz, Leonid Karlinsky, Rog ´erio Schmidt Feris, Raja Giryes, and Alex M. Bronstein. Baby steps towards few-shot learning with multiple semantics. arXiv preprint arXiv:1906.01905, 2019. 2

    1906 arXiv

  99. [107]

    Zero- shot sketch-image hashing

    Yuming Shen, Li Liu, Fumin Shen, and Ling Shao. Zero- shot sketch-image hashing. In CVPR, 2018. 2 11

    2018

  100. [108]

    Prototypi- cal networks for few-shot learning

    Jake Snell, Kevin Swersky, and Richard Zemel. Prototypi- cal networks for few-shot learning. In NeurIPS, 2017. 2, 3, 4, 5, 14

    2017

  101. [109]

    Soria, E

    X. Soria, E. Riba, and A. Sappa. Dense extreme inception network: Towards a robust cnn model for edge detection. In WACV, 2020. 2, 16

    2020

  102. [110]

    Pixel difference networks for efficient edge detection

    Zhuo Su, Wenzhe Liu, Zitong Yu, Dewen Hu, Qing Liao, Qi Tian, Matti Pietik ¨ainen, and Li Liu. Pixel difference networks for efficient edge detection. In ICCV, 2021. 3, 5, 6, 7, 8, 16, 17

    2021

  103. [111]

    Deep high-resolution representation learning for human pose es- timation

    Ke Sun, Bin Xiao, Dong Liu, and Jingdong Wang. Deep high-resolution representation learning for human pose es- timation. In CVPR, 2019. 3

    2019

  104. [112]

    Uniap: Towards universal animal percep- tion in vision via few-shot learning

    Meiqi Sun, Zhonghan Zhao, Wenhao Chai, Hanjun Luo, Shidong Cao, Yanting Zhang, Jenq-Neng Hwang, and Gaoang Wang. Uniap: Towards universal animal percep- tion in vision via few-shot learning. In AAAI, 2024. 3, 18

    2024

  105. [113]

    Deep convolu- tional network cascade for facial point detection

    Yi Sun, Xiaogang Wang, and Xiaoou Tang. Deep convolu- tional network cascade for facial point detection. In CVPR,

  106. [114]

    Learning to compare: Relation network for few-shot learning

    Flood Sung, Yongxin Yang, Li Zhang, Tao Xiang, Philip HS Torr, and Timothy M Hospedales. Learning to compare: Relation network for few-shot learning. InCVPR,

  107. [115]

    A prototype-oriented framework for unsupervised domain adaptation

    Korawat Tanwisuth, Xinjie Fan, Huangjie Zheng, Shu- jian Zhang, Hao Zhang, Bo Chen, and Mingyuan Zhou. A prototype-oriented framework for unsupervised domain adaptation. In NeurIPS, 2021. 2, 3, 5, 6, 8, 14, 17

    2021

  108. [116]

    Unsupervised learning of landmarks by descriptor vector exchange

    James Thewlis, Samuel Albanie, Hakan Bilen, and Andrea Vedaldi. Unsupervised learning of landmarks by descriptor vector exchange. In ICCV, 2019. 1, 2, 3

    2019

  109. [117]

    Deeppose: Hu- man pose estimation via deep neural networks

    Alexander Toshev and Christian Szegedy. Deeppose: Hu- man pose estimation via deep neural networks. In CVPR,

  110. [118]

    Matching networks for one shot learning

    Oriol Vinyals, Charles Blundell, Timothy Lillicrap, Daan Wierstra, et al. Matching networks for one shot learning. In NeurIPS, 2016. 2

    2016

  111. [119]

    Sketchembednet: Learning novel concepts by imitating drawings

    Alexander Wang, Mengye Ren, and Richard Zemel. Sketchembednet: Learning novel concepts by imitating drawings. In ICML, 2021. 2

    2021

  112. [120]

    What makes training multi-modal classification networks hard? In CVPR, 2020

    Weiyao Wang, Du Tran, and Matt Feiszli. What makes training multi-modal classification networks hard? In CVPR, 2020. 3

    2020

  113. [121]

    Generalizing from a few examples: A survey on few-shot learning

    Yaqing Wang, Quanming Yao, James T Kwok, and Li- onel M Ni. Generalizing from a few examples: A survey on few-shot learning. ACM CSur, 2020. 2

    2020

  114. [122]

    Low-shot learning from imaginary data

    Yu-Xiong Wang, Ross Girshick, Martial Hebert, and Bharath Hariharan. Low-shot learning from imaginary data. In CVPR, 2018. 2

    2018

  115. [123]

    Few-shot model adaptation for customized facial landmark detection, segmentation, stylization and shadow removal

    Zhen Wei, Bingkun Liu, Weinong Wang, and Yu-Wing Tai. Few-shot model adaptation for customized facial landmark detection, segmentation, stylization and shadow removal. arXiv preprint arXiv:2104.09457, 2021. 1, 3

    2021 arXiv

  116. [124]

    Meta-learning: Learning to learn fast

    Lilian Weng. Meta-learning: Learning to learn fast. lilian- weng.github.io, 2018. 2

    2018

  117. [125]

    Memory networks

    Jason Weston, Sumit Chopra, and Antoine Bordes. Memory networks. arXiv preprint arXiv:1410.3916, 2014. 2

    2014 arXiv

  118. [126]

    Back to 3d: Few-shot 3d keypoint detection with back- projected 2d features

    Thomas Wimmer, Peter Wonka, and Maks Ovsjanikov. Back to 3d: Few-shot 3d keypoint detection with back- projected 2d features. In CVPR, 2024. 3

    2024

  119. [127]

    Stacked cross re- finement network for edge-aware salient object detection

    Zhe Wu, Li Su, and Qingming Huang. Stacked cross re- finement network for edge-aware salient object detection. In ICCV, 2019. 5

    2019

  120. [128]

    Exploit- ing aliasing for manga restoration

    Minshan Xie, Menghan Xia, and Tien-Tsin Wong. Exploit- ing aliasing for manga restoration. In CVPR, 2021. 2

    2021

  121. [129]

    Holistically-nested edge de- tection

    Saining Xie and Zhuowen Tu. Holistically-nested edge de- tection. In ICCV, 2015. 3, 5, 8, 16

    2015

  122. [130]

    Oreshkin, and Pedro H

    Chen Xing, Negar Rostamzadeh, Boris N. Oreshkin, and Pedro H. O. Pinheiro. Adaptive cross-modal few-shot learn- ing. In NeurIPS, 2019. 2, 3

    2019

  123. [131]

    Deep learning for free-hand sketch: A survey

    Peng Xu, Timothy M Hospedales, Qiyue Yin, Yi-Zhe Song, Tao Xiang, and Liang Wang. Deep learning for free-hand sketch: A survey. IEEE TPAMI, 2022. 2

    2022

  124. [132]

    Interactive liquid splash modeling by user sketches

    Guowei Yan, Zhili Chen, Jimei Yang, and Huamin Wang. Interactive liquid splash modeling by user sketches. ACM TOG, 2020. 2

    2020

  125. [133]

    Object pose estimation with statistical guarantees: Conformal keypoint detection and geometric uncertainty propagation

    Heng Yang and Marco Pavone. Object pose estimation with statistical guarantees: Conformal keypoint detection and geometric uncertainty propagation. In CVPR, 2023. 1, 2

    2023

  126. [134]

    Deep plastic surgery: Robust and controllable image editing with human-drawn sketches

    Shuai Yang, Zhangyang Wang, Jiaying Liu, and Zongming Guo. Deep plastic surgery: Robust and controllable image editing with human-drawn sketches. In ECCV, 2020. 2

    2020

  127. [135]

    Trans- pose: Keypoint localization via transformer

    Sen Yang, Zhibin Quan, Mu Nie, and Wankou Yang. Trans- pose: Keypoint localization via transformer. InICCV, 2021. 1, 2

    2021

  128. [136]

    Automated relational meta-learning

    Huaxiu Yao, Xian Wu, Zhiqiang Tao, Yaliang Li, Bolin Ding, Rui rui Li, and Zhenhui Li. Automated relational meta-learning. In ICLR, 2020. 2

    2020

  129. [137]

    One-shot medical landmark detection

    Qingsong Yao, Quan Quan, Li Xiao, and S Kevin Zhou. One-shot medical landmark detection. In MICCAI, 2021. 3

    2021

  130. [138]

    A zero-shot framework for sketch based image retrieval

    Sasi Kiran Yelamarthi, Shiva Krishna Reddy, Ashish Mishra, and Anurag Mittal. A zero-shot framework for sketch based image retrieval. In ECCV, 2018. 2

    2018

  131. [139]

    Adversarial meta-learning

    Chengxiang Yin, Jian Tang, Zhiyuan Xu, and Yanzhi Wang. Adversarial meta-learning. arXiv preprint arXiv:1806.03316, 2018. 2

    2018 arXiv

  132. [140]

    Bayesian model- agnostic meta-learning

    Jaesik Yoon, Taesup Kim, Ousmane Dia, Sungwoong Kim, Yoshua Bengio, and Sungjin Ahn. Bayesian model- agnostic meta-learning. In NeurIPS, 2018. 2

    2018

  133. [141]

    Tapnet: Neural network augmented with task-adaptive projection for few-shot learning

    Sung Whan Yoon, Jun Seo, and Jaekyun Moon. Tapnet: Neural network augmented with task-adaptive projection for few-shot learning. In ICML, 2019. 2

    2019

  134. [142]

    Sketch me that shoe

    Qian Yu, Feng Liu, Yi-Zhe Song, Tao Xiang, Timothy M Hospedales, and Chen-Change Loy. Sketch me that shoe. In CVPR, 2016. 2, 3, 17

    2016

  135. [143]

    Sketch2model: View-aware 3d modeling from single free- hand sketches

    Song-Hai Zhang, Yuan-Chen Guo, and Qing-Wen Gu. Sketch2model: View-aware 3d modeling from single free- hand sketches. In CVPR, 2021. 2

    2021

  136. [144]

    Deep supervised cross-modal retrieval

    Liangli Zhen, Peng Hu, Xu Wang, and Dezhong Peng. Deep supervised cross-modal retrieval. In CVPR, 2019. 3

    2019

  137. [145]

    Learning to learn image classifiers with visual analogy

    Linjun Zhou, Peng Cui, Shiqiang Yang, Wenwu Zhu, and Qi Tian. Learning to learn image classifiers with visual analogy. In CVPR, 2019. 2 12

    2019

  138. [146]

    Ob- jects as points

    Xingyi Zhou, Dequan Wang, and Philipp Kr ¨ahenb¨uhl. Ob- jects as points. arXiv preprint arXiv:1904.07850, 2019. 2

    1904 arXiv

  139. [147]

    Bottom-up object detection by grouping extreme and center points

    Xingyi Zhou, Jiacheng Zhuo, and Philipp Krahenbuhl. Bottom-up object detection by grouping extreme and center points. In CVPR, 2019. 2

    2019

  140. [148]

    Zintgraf, Kyriacos Shiarlis, Vitaly Kurin, Katja Hofmann, and Shimon Whiteson

    Luisa M. Zintgraf, Kyriacos Shiarlis, Vitaly Kurin, Katja Hofmann, and Shimon Whiteson. Fast context adaptation via meta-learning. In ICML, 2019. 2 13 Doodle Your Keypoints: Sketch-Based Few-Shot Keypoint Detection Supplementary Material

    2019

  141. [149]

    Insights on Framework Design 6.1. Transport Loss for Keypoint Localization Background: Prototypical domain adaptation [115] is particularly curated for classification that differs from key- point detection [67] significantly at the task level. Thus, to adapt the same for the s...

  142. [150]

    Design Analysis of De-stylization Network The de-stylization network Z disentangles style and spar- sity from the keypoint embeddings using attentional global and local context fusion [25]. However, to understand the role of global context realized by the global pooled vector ...

  143. [151]

    + [109] + [129] 36.87 12.46

  144. [152]

    + [56] + [14] 37.33 12.98

  145. [153]

    + [60] + [109] 38.38 13.45

  146. [154]

    Performance of the proposed framework with various edge detection algorithms used for generating synthetic sketch data for training

    + [60] + [56] 38.61 13.87 Ours ([110] + [129] + [14]) 39.00 14.42 Table 4. Performance of the proposed framework with various edge detection algorithms used for generating synthetic sketch data for training. Performances (PCK@0.1) are measured on both datasets [15, 79] for nov...

  147. [155]

    As we treat edgemaps as synthetic sketch data, we performed in-depth experimentation with the different edge detection algo- rithms as mentioned in Sec

    Choosing Edge Detection Algorithms The choice of edge detection algorithm is particularly crucial for our problem, as it directly connects with the training data for the proposed framework. As we treat edgemaps as synthetic sketch data, we performed in-depth experimentation wi...

  148. [156]

    3 and 4, and a few sample visualizations of support edgemap [110] and detec- tion with ground-truth on query photos [15] are given in Fig

    Empirical Study with Free-Hand Sketches The proposed framework is entirely trained with edgemaps or synthetic sketches as given in Secs. 3 and 4, and a few sample visualizations of support edgemap [110] and detec- tion with ground-truth on query photos [15] are given in Fig. 1...

  149. [157]

    [67] solves

    Experiments with Support Photos Apart from using sketches or edgemaps as support, we also experiment with photos as support, solving the simple few- shot keypoint detection problem Lu et al. [67] solves. This experimentation was conducted to prove the robustness of the propose...

  150. [158]

    3.3) disen- tangles the style-invariant features

    Style Diversity in Real Sketches We simulate user sketch styles [100] with different edgemaps as the de-stylization network (Sec. 3.3) disen- tangles the style-invariant features. In order to under- stand its style-invariance and generalization capability to real sketches, we ...

  151. [159]

    Challenges with Additional Modalities Our few-shot framework is particularly curated for sketch- photo cross-modal learning, and is vastly different from traditional sketch research [7, 8, 20], as careful address- ing of the domain shift is well-observed in sketch-photo cross-...

  152. [160]

    Additional Comparisons Although few-shot keypoint learning [33, 69, 112] has been around for some time, only a few state-of-the-art methods are suitable for comparison to the proposed method. Our framework follows FSKD [67] closely, and we compare Class Keypoints Methods PCK@0...

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.