Pith. sign in

REVIEW 3 major objections 5 minor 48 references

Unsupervised Part Discovery via Descriptor-Based Masked Image Restoration with Optimized Constraints

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read MPAE discovers meaningful object parts with no labels: filling randomly masked patches with learned part descriptors and restoring the image aligns descriptor semantics with real part shapes, and the method outperforms baselines on…

desk verdict Genuinely new mechanism for unsupervised part discovery, but the paper's own ablation shows the reported NMI/ARI metrics reward a degenerate foreground-segmentation solution, so the quantitative claims need a credibility fix. read the letter →

arxiv 2507.11985 v1 pith:EJ3ULHWT submitted 2025-07-16 cs.CV

Jiahao Xia , Yike Wu , Wenjian Huang , Jianguo Zhang , Jian Zhang This is my paper · ORCID
classification cs.CV
keywords unsupervisedpartdiscoverymaskedautoencoderdescriptorssimilaritymapssegmentationself-supervisedVisionTransformerperceptuallosspresenceconstraints
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

MPAE (Masked Part Autoencoder) is a training paradigm for discovering object parts without any part or category labels. Instead of reconstructing the whole image directly from part descriptors, it randomly masks patches, fills the masked regions with learned part descriptors according to similarity maps, and then restores the image from that filled feature map. The restoration step is what aligns the high-level semantics of the descriptors with actual part shapes, because the visible patches supply low-level appearance cues such as boundaries and textures. With relaxed presence constraints that only require each part to appear somewhere in a small group of images, the model can handle multi-category data and occlusion. The paper reports that this yields masks that closely match object parts on PartImageNet, CUB, and CelebA, outperforming both unsupervised and weakly supervised baselines.

What carries the argument

The central object is the Masked Part Autoencoder and its matching block. A frozen self-supervised Vision Transformer plus a trainable $1 \times 1$ convolution produces a dense feature map $F$; the model learns $K+1$ part descriptors $D$ (one per foreground part plus background); softmax similarity $P_{i,j,k}$ scores each pixel against each descriptor; masked positions of the filled feature map $R$ are set to $\sum_{k=1}^{K+1} P_{i,j,k} D_k$, while visible positions keep the unmasked patch features $F^U$. Restoration with an L1 term plus a perceptual term then pulls descriptors and visible features into shared latent clusters. Three constraints stabilize this: a presence constraint (each part appears somewhere in a mini-group, background near image borders), a semantic consistency constraint with additive angular margin on descriptor-to-region cosine similarity, and a distribution constraint combining total variation and entropy to sharpen part boundaries.

What would settle it

Train MPAE with a decoder strong enough to reconstruct masked patches almost perfectly from visible context alone, for example by using a much deeper decoder or a very low masking ratio, and then measure whether the similarity maps still match annotated part boundaries. If reconstruction stays accurate while the similarity maps no longer track part shapes, the claimed implicit alignment through restoration is not the driving mechanism.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that masked image restoration can function as a part-alignment mechanism: by using learned part descriptors to fill masked patches and then restoring the image, the model implicitly clusters the filled descriptor features together with unmasked patch features inside each part region, while low-level appearance from visible patches guides the descriptors to the correct part boundaries. The resulting similarity maps can be read directly as pixel-level part masks. The paper further argues that looser constraints, which allow parts to be absent from individual images, let the same descriptors transfer across categories and handle occlusion, and it supports this with experiments showing improved normalized mutual information and adjusted Rand index over existing methods on four benchmarks.

Load-bearing premise

The load-bearing premise is that filling masked patches with part descriptors and restoring the image will, by itself, pull descriptors and unmasked patch features into shared latent clusters whose similarity maps track part shapes; the paper's ablation shows this alignment is fragile, since removing the perceptual term drops normalized mutual information (NMI) from 55.10 to 19.65 on the multi-category segmentation benchmark.

Editorial extensions

If this is right

  • With no labels, MPAE can produce part-level masks that follow real object boundaries, which is a prerequisite for part-level features in downstream tasks such as fine-grained recognition and person search.
  • Because parts may be absent from any single image, the same training procedure transfers across categories, allowing the model to share similar parts such as wheels, heads, or hulls across different object classes.
  • The method is designed to handle occlusion, since it learns which parts are present rather than assuming all parts appear in every image.
  • The optimal masking ratio around 90 percent indicates that the alignment mechanism depends on forcing most of the reconstruction through part descriptors rather than through visible context alone.

Reading between the lines

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

  • A testable extension follows from the masking-ratio curve: if descriptor alignment is driven by restoration pressure, then an annealing schedule that raises the masking ratio during training could improve convergence and final part quality, though the paper does not test this.
  • The role of the perceptual loss suggests the method's success is tied to structural supervision in pixel space; one could test whether a latent-space perceptual loss, or a stronger structural criterion, changes how tightly similarity maps follow boundaries.
  • The mini-group presence constraint is an implicit batch-level prior; the paper's ablation shows very large groups fragment rare regions into spurious parts, so a principled choice of group size could be connected to part frequency statistics across categories.
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

3 major / 5 minor

Summary. This paper proposes MPAE, a masked autoencoder variant for unsupervised part discovery. The model learns K+1 part descriptors from a trainable ViT, extracts dense features from a frozen pretrained ViT, and uses the similarity between descriptors and local features to fill masked patches in an image restoration objective. Several auxiliary losses are introduced: a presence loss (Lf and Lb), a semantic consistency loss (Ls), and a distribution loss (Ld). The paper reports NMI/ARI comparisons on PartImageNet, CUB, and CelebA, plus extensive ablations over masking ratios, hyperparameters, backbones, and loss components. The code is provided.

Significance. The idea of using masked restoration with part descriptors as fill tokens is interesting, and the qualitative results suggest plausible part masks across categories and scenarios. The paper ships code and includes extensive ablations, which is a strength. If the quantitative claims held, this would be a useful step toward unsupervised part discovery. However, the evaluation issues detailed below mean that the numerical superiority over prior work is not currently established, and the paper's own ablations indicate that the primary metric may reward a degenerate foreground/background solution.

major comments (3)
  1. [Section 4.4, Table 4] The paper's primary evidence is NMI/ARI, yet its own ablation shows these metrics reward a degenerate solution. Removing Ls increases NMI/ARI at K=8 (55.69/76.54 vs 33.51/65.05) and K=25 (57.70/76.97 vs 37.22/68.04), while the text states the model without Ls 'degrades into a foreground segmentation model' and Appendix B.8 reports only 3.94 foreground parts per image. Thus the higher NMI/ARI of the full model at K=50 (55.10/73.52) could in principle be driven by coarse foreground/background separation rather than meaningful part structure. The comparisons in Tables 1-3 are therefore contaminated: they do not establish that MPAE discovers parts better than a simple foreground segmentation model. Please add baselines that explicitly evaluate part-level quality (e.g., part-aware IoU or semantic part consistency) and report the number of active parts per image for all methods.
  2. [Table 1 and Table 6] The comparison with Xia et al. is at a different cluster count (K=4Nc, i.e., 436 for OOD and 632 for Segmentation) versus MPAE at K=8/25/50. NMI and ARI are known to depend on the number of clusters, so these numbers are not directly comparable. Notably, Table 6 shows that changing MPAE's K from 50 to 4Nc reduces NMI from 55.10 to 47.59 and ARI from 73.52 to 70.71, which is a large drop. Please either evaluate MPAE at the same K as Xia et al., or evaluate Xia et al. at the same K as MPAE, and discuss the cluster-count dependence.
  3. [Section 3.1.1 and Appendix B.1] The paper asserts that the restoration objective 'implicitly clusters' the filled descriptors and unmasked patch features, and that this is the mechanism that aligns descriptors with part shapes. The only direct evidence is the large drop when the VGG-19 structural penalty is removed (NMI 55.10 to 19.65, ARI 73.52 to 49.72 in Table 7). This shows the component matters but does not show that the clustering actually occurs as described. Please provide direct evidence, such as an analysis of the latent space distribution of filled versus unmasked features, or a controlled experiment that varies restoration difficulty, or temper the mechanistic claim to what is demonstrated.
minor comments (5)
  1. [Section 1 and Section 4.1] Typos: 'inpus' should be 'inputs'; 'datasests' should be 'datasets'.
  2. [Contributions list] The phrase 'as followings' should be 'as follows'.
  3. [Equation (7)] The typesetting of the ArcFace-style loss is garbled (e.g., 'M kes' appears to be M_k * exp(...)). Please rewrite with clear mathematical notation and define all symbols in one place.
  4. [Section 3.3] The term 'PartFormer' is used without a formal definition; earlier in Section 3.1.1 it is mentioned in passing. Please introduce the term when the architecture is first described.
  5. [Section 4.4, masking ratio paragraph] The sentence 'a higher r results in more features in the filled feature map R comes from part descriptors D' has a subject-verb agreement problem and should be rephrased.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: MPAE's central claims are tested against external benchmarks with standard training losses; the cited self-work is design inheritance, not a load-bearing derivation.

full rationale

The paper's derivation chain is not circular at the equation level. The training objective, Eq. (3), is a masked image restoration loss combining L1 reconstruction with a VGG-19 perceptual penalty, and the auxiliary constraints in Eqs. (4)-(11) are standard presence, semantic-consistency, total-variation, and entropy losses. None of these losses contains the evaluation metrics NMI, ARI, or NME; the reported numbers are computed against external human annotations on PartImageNet, CUB, and CelebA. The part descriptors D and similarity maps P are learned parameters and data-dependent outputs, not fitted values that are later renamed as predictions. The self-citations to the authors' prior TPAMI work [43] are limited to architectural choices and hyperparameters: 'as [43], we replace the class embedding of a classical ViT with K + 1 embeddings' and 'Referring to [43], we set s and m to 20 and 0.5.' These borrowings are implementation details, and the paper actively compares against [43] and reports better numbers, so the conclusion does not rest on an unverified self-citation. The ablation in Table 4, where removing Ls improves NMI/ARI while the paper states the model 'degrades into a foreground segmentation model,' is a measurement-validity concern rather than circularity: it says the metric can reward coarse foreground splits, but it does not show that the full model's result is equivalent by construction to its inputs. The mechanism claim in Section 3.1.1 that restoration 'implicitly clusters FU and D within the same part region together in latent space' is an empirical assumption supported by ablations, not a definitional equivalence. No uniqueness theorem, fitted-input-as-prediction, or ansatz-smuggled-via-citation pattern is present.

Assumptions & free parameters 7 free parameters · 4 assumptions · 0 invented entities

The central claim rests on a large pretrained model (DINOv2), several hand-tuned hyperparameters, and an empirically assumed implicit-clustering mechanism. No new physical or conceptual entities are postulated: part descriptors are a representation inherited from the prior literature, so the invented-entities list is empty.

free parameters (7)
  • K (number of parts) = 8, 25, 50 (PartImageNet); 4, 8, 16 (CUB, CelebA)
    The desired number of discovered parts is chosen by hand per dataset and strongly affects NMI/ARI; no criterion derives K from data.
  • Masking ratio r = 0.90
    Tuned via sweep from 50% to 100% on PartImageNet Segmentation (Table 5); optimal 90%.
  • Mini-group size G = 64 (multi-category), 8 (single-category)
    Tuned per dataset type in Appendix B.7; too small or too large degrades performance.
  • Loss weights lambda_d, lambda_p, lambda_s = 0.5, 1.0, 0.25
    Tuned on PartImageNet Segmentation (Appendix B.4); kept fixed across datasets.
  • ArcFace scale s and margin m = 20 and 0.5
    Inherited from the authors' prior work [43] (Appendix A.2); influence the semantic consistency loss.
  • Presence threshold for M_k = 0.001
    Ad hoc threshold in Eq. (7) deciding whether a part is present in an image.
  • MPAE encoder/decoder layer counts = 2 layers each
    Selected by ablation in Appendices B.2 and B.3.
assumptions (4)
  • domain assumption The softmax similarity map P between learned descriptors and pretrained features can serve as a pixel-level part mask after training.
    Used in Eq. (1) and at inference (Section 3.3); the entire evaluation treats P as the part segmentation output.
  • domain assumption Masked restoration implicitly clusters descriptor-filled features with unmasked features of the same part in latent space.
    Section 3.1.1 claims this without proof; supported only by ablation (Appendix B.1) and a citation to masked autoencoder theory [45].
  • domain assumption Each mini-group of G images contains all K foreground parts and background appears near image boundaries.
    This is the premise of Lf (Eq. 4) and Lb (Eq. 5); it can fail for rare parts or unusual compositions, and G is tuned to make it plausible.
  • domain assumption NMI/ARI between predicted clusters and annotated part masks measures part discovery quality.
    The paper's own Table 4 shows a model without Ls scores higher on these metrics while being labeled as a failure, so the metric is not a faithful proxy for the stated goal.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Unsupervised Part Discovery via Descriptor-Based Masked Image Restoration with Optimized Constraints." pith.science (2026). https://pith.science/paper/EJ3ULHWT

@misc{pith2026250711985,
  author       = {Pith},
  title        = {Pith review of: Unsupervised Part Discovery via Descriptor-Based Masked Image Restoration with Optimized Constraints},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EJ3ULHWT}},
  note         = {Machine review of arXiv:2507.11985}
}
read the original abstract

Part-level features are crucial for image understanding, but few studies focus on them because of the lack of fine-grained labels. Although unsupervised part discovery can eliminate the reliance on labels, most of them cannot maintain robustness across various categories and scenarios, which restricts their application range. To overcome this limitation, we present a more effective paradigm for unsupervised part discovery, named Masked Part Autoencoder (MPAE). It first learns part descriptors as well as a feature map from the inputs and produces patch features from a masked version of the original images. Then, the masked regions are filled with the learned part descriptors based on the similarity between the local features and descriptors. By restoring these masked patches using the part descriptors, they become better aligned with their part shapes, guided by appearance features from unmasked patches. Finally, MPAE robustly discovers meaningful parts that closely match the actual object shapes, even in complex scenarios. Moreover, several looser yet more effective constraints are proposed to enable MPAE to identify the presence of parts across various scenarios and categories in an unsupervised manner. This provides the foundation for addressing challenges posed by occlusion and for exploring part similarity across multiple categories. Extensive experiments demonstrate that our method robustly discovers meaningful parts across various categories and scenarios. The code is available at the project https://github.com/Jiahao-UTS/MPAE.

Figures

Figures reproduced from arXiv: 2507.11985 by the authors.

Figure 1
Figure 1. MPAE uses similarity maps between local image fea [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Training and inference pipelines of MPAE. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Some qualitative results of MPAE (ours) and other state [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Some visualized part discovery results of MPAE (ours) and state-of-the-art methods on CUB dataset ( [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Qualitative results from the ablation studies of the loss [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Some qualitative results of image restoration and part discovery predicted by the MPAE trained with/without structural difference [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]
Figure 7
Figure 7. Figure 7: Parts unsupervisedly discovered by MPAE across multiple categories on PartImageNet Segmentation ( [PITH_FULL_IMAGE:figures/full_fig_p015_7.png]
Figure 8
Figure 8. Figure 8: Pixel-level masks of discovered parts predicted by MPAE and their corresponding attention maps on PartImageNet Segmentation [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]
Figure 9
Figure 9. Figure 9: Examples of unsupervised part discovery results on PartImageNet OOD dataset predicted by MPAE in the setting of [PITH_FULL_IMAGE:figures/full_fig_p016_9.png]
Figure 10
Figure 10. Figure 10: Examples of unsupervised part discovery results on PartImageNet Segmentation dataset predicted by MPAE in the setting of [PITH_FULL_IMAGE:figures/full_fig_p016_10.png]
Figure 11
Figure 11. Figure 11: Examples of unsupervised part discovery results on CUB dataset predicted by MPAE in the setting of [PITH_FULL_IMAGE:figures/full_fig_p017_11.png]
Figure 12
Figure 12. Figure 12: Examples of unsupervised part discovery results on CelebA dataset predicted by MPAE in the setting of [PITH_FULL_IMAGE:figures/full_fig_p017_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

48 extracted references · 47 canonical work pages

  1. [1]

    Deep vit features as dense visual descriptors

    Shir Amir, Yossi Gandelsman, Shai Bagon, and Tali Dekel. Deep vit features as dense visual descriptors. InECCV Work- shops, 2022. 1, 3, 6, 7, 13

    work page 2022

  2. [2]

    Pdiscoformer: Relaxing part discovery constraints with vision transformers

    Ananthu Aniraj, Cassio F Dantas, Dino Ienco, and Diego Marcos. Pdiscoformer: Relaxing part discovery constraints with vision transformers. In ECCV, 2024. 1, 2, 4, 6, 7, 8, 11

    work page 2024

  3. [3]

    Unsupervised learn- ing of visual features by contrasting cluster assignments

    Mathilde Caron, Ishan Misra, Julien Mairal, Priya Goyal, Pi- otr Bojanowski, and Armand Joulin. Unsupervised learn- ing of visual features by contrasting cluster assignments. In NIPS, pages 9912–9924. Curran Associates, Inc., 2020. 2

    work page 2020

  4. [4]

    Emerg- ing properties in self-supervised vision transformers

    Mathilde Caron, Hugo Touvron, Ishan Misra, Herv ´e Jegou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerg- ing properties in self-supervised vision transformers. In ICCV, pages 9630–9640, 2021. 1, 2, 13

    work page 2021

  5. [5]

    An empiri- cal study of training self-supervised vision transformers

    Xinlei Chen, Saining Xie, and Kaiming He. An empiri- cal study of training self-supervised vision transformers. In ICCV, pages 9620–9629, 2021. 2

    work page 2021

  6. [6]

    Unsupervised part discovery from con- trastive reconstruction

    Subhabrata Choudhury, Iro Laina, Christian Rupprecht, and Andrea Vedaldi. Unsupervised part discovery from con- trastive reconstruction. In NIPS, pages 28104–28118, 2021. 1, 3, 7

    work page 2021

  7. [7]

    Deep feature factorization for concept discovery

    Edo Collins, Radhakrishna Achanta, and Sabine S ¨usstrunk. Deep feature factorization for concept discovery. In ECCV, pages 352–368, 2018. 3

    work page 2018

  8. [8]

    Vision transformers need registers

    Timoth ´ee Darcet, Maxime Oquab, Julien Mairal, and Piotr Bojanowski. Vision transformers need registers. In ICLR.,

    work page

Show all 48 references
  1. [9]

    Arcface: Additive angular mar- gin loss for deep face recognition

    Jiankang Deng, Jia Guo, Jing Yang, Niannan Xue, Irene Kot- sia, and Stefanos Zafeiriou. Arcface: Additive angular mar- gin loss for deep face recognition. IEEE TPAMI, 44(10): 5962–5979, 2022. 5

    2022

  2. [10]

    BERT: Pre-training of deep bidirectional trans- formers for language understanding

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep bidirectional trans- formers for language understanding. InNAACL, pages 4171– 4186, 2019. 2

    2019

  3. [11]

    Adam: A method for stochastic optimization

    Kingma Diederik and Ba Jimmy. Adam: A method for stochastic optimization. In ICLR, 2015. 11

    2015

  4. [12]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition ...

    2021

  5. [13]

    Un- supervised co-part segmentation through assembly

    Qingzhe Gao, Bin Wang, Libin Liu, and Baoquan Chen. Un- supervised co-part segmentation through assembly. InICML, pages 3576–3586, 2021. 1, 2, 3

    2021

  6. [14]

    Bootstrap your own latent - a new approach to self-supervised learning

    Jean-Bastien Grill, Florian Strub, Florent Altch ´e, Corentin Tallec, Pierre Richemond, Elena Buchatskaya, Carl Doersch, Bernardo Avila Pires, Zhaohan Guo, Mohammad Ghesh- laghi Azar, Bilal Piot, koray kavukcuoglu, Remi Munos, and Michal Valko. Bootstrap your own latent - a ne...

  7. [15]

    Siamese masked autoencoders

    Agrim Gupta, Jiajun Wu, Jia Deng, and Fei-Fei Li. Siamese masked autoencoders. In NIPS, pages 40676–40693. Curran Associates, Inc., 2023. 2

    2023

  8. [16]

    Partimagenet: A large, high- quality dataset of parts

    Ju He, Shuo Yang, Shaokang Yang, Adam Kortylewski, Xi- aoding Yuan, Jie-Neng Chen, Shuai Liu, Cheng Yang, Qi- hang Yu, and Alan Yuille. Partimagenet: A large, high- quality dataset of parts. In ECCV, pages 128–145, 2022. 5, 11

    2022

  9. [18]

    Momentum contrast for unsupervised visual rep- resentation learning

    Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum contrast for unsupervised visual rep- resentation learning. In CVPR, pages 9726–9735, 2020. 2

    2020

  10. [19]

    Masked autoencoders are scalable vision learners

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Doll´ar, and Ross Girshick. Masked autoencoders are scalable vision learners. In CVPR, pages 15979–15988, 2022. 2, 3, 4

    2022

  11. [20]

    Ganseg: Learning to segment by unsupervised hierarchical image generation

    Xingzhe He, Bastian Wandt, and Helge Rhodin. Ganseg: Learning to segment by unsupervised hierarchical image generation. In CVPR, pages 1215–1225, 2022. 1, 2, 7

    2022

  12. [21]

    Interpretable and accurate fine- grained recognition via region grouping

    Zixuan Huang and Yin Li. Interpretable and accurate fine- grained recognition via region grouping. In CVPR, pages 8659–8669, 2020. 2, 6, 7, 11

    2020

  13. [22]

    Scops: Self-supervised co-part segmentation

    Wei-Chih Hung, Varun Jampani, Sifei Liu, Pavlo Molchanov, Ming-Hsuan Yang, and Jan Kautz. Scops: Self-supervised co-part segmentation. In CVPR, pages 869–878, 2019. 1, 3, 5, 8, 11

    2019

  14. [23]

    Perceptual losses for real-time style transfer and super-resolution

    Justin Johnson, Alexandre Alahi, and Li Fei-Fei. Perceptual losses for real-time style transfer and super-resolution. In ECCV, pages 694–711, 2016. 4

    2016

  15. [24]

    Berg, Wan-Yen Lo, Piotr Doll ´ar, and Ross Girshick

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer White- head, Alexander C. Berg, Wan-Yen Lo, Piotr Doll ´ar, and Ross Girshick. Segment anything. In ICCV, pages 3992– 4003, 2023. 14

    2023

  16. [25]

    Unsupervised part segmentation through disentangling ap- pearance and shape

    Shilong Liu, Lei Zhang, Xiao Yang, Hang Su, and Jun Zhu. Unsupervised part segmentation through disentangling ap- pearance and shape. In CVPR, pages 8351–8360, 2021. 1, 2, 3

    2021

  17. [26]

    Deep learning face attributes in the wild

    Ziwei Liu, Ping Luo, Xiaogang Wang, and Xiaoou Tang. Deep learning face attributes in the wild. In ICCV, pages 3730–3738, 2015. 5, 11

    2015

  18. [27]

    Animal kingdom: A large and diverse 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 diverse dataset for animal behavior understanding. InCVPR, pages 19001–19012, 2022. 1

    2022

  19. [28]

    Maxime Oquab, Timoth ´ee Darcet, Th´eo Moutakanni, Huy V . V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel HAZIZA, Francisco Massa, Alaaeldin El-Nouby, Mido Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michael ...

    2024

  20. [29]

    Plot: Text-based person search with part slot attention for corresponding part discovery

    Jicheol Park, Dongwon Kim, Boseung Jeong, and Suha Kwak. Plot: Text-based person search with part slot attention for corresponding part discovery. In ECCV, pages 474–490,

  21. [30]

    Improving language understanding by generative pre-training

    Alec Radford, Karthik Narasimhan, Tim Salimans, and Ilya Sutskever. Improving language understanding by generative pre-training. OpenAI blog, 2018. 2

    2018

  22. [31]

    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, pages 8748–8763, 2021. 14

    2021

  23. [32]

    Rudin, Stanley Osher, and Emad Fatemi

    Leonid I. Rudin, Stanley Osher, and Emad Fatemi. Nonlinear total variation based noise removal algorithms. Physica D: Nonlinear Phenomena, 60(1):259–268, 1992. 5

    1992

  24. [33]

    Particle: Part discov- ery and contrastive learning for fine-grained recognition

    Oindrila Saha and Subhransu Maji. Particle: Part discov- ery and contrastive learning for fine-grained recognition. In ICCV Workshops, pages 167–176, 2023. 2

    2023

  25. [34]

    Jianbo Shi and J. Malik. Normalized cuts and image seg- mentation. IEEE TPAMI, 22(8):888–905, 2000. 2

    2000

  26. [35]

    Motion- supervised co-part segmentation

    Aliaksandr Siarohin, Subhankar Roy, St ´ephane Lathuili `ere, Sergey Tulyakov, Elisa Ricci, and Nicu Sebe. Motion- supervised co-part segmentation. In ICPR, pages 9650– 9657, 2021. 1, 3

    2021

  27. [36]

    Very deep convo- lutional networks for large-scale image recognition

    Karen Simonyan and Andrew Zisserman. Very deep convo- lutional networks for large-scale image recognition. InICLR,

  28. [37]

    Unsu- pervised learning of object landmarks by factorized spatial embeddings

    James Thewlis, Hakan Bilen, and Andrea Vedaldi. Unsu- pervised learning of object landmarks by factorized spatial embeddings. In ICCV, pages 3229–3238, 2017. 3

    2017

  29. [38]

    van der Klis, S

    R. van der Klis, S. Alaniz, M. Mancini, C. F. Dantas, D. Ienco, Z. Akata, and D. Marcos. Pdisconet: Semantically consistent part discovery for fine-grained recognition. In ICCV, pages 1866–1876, 2023. 1, 2, 6, 7, 11

    2023

  30. [39]

    C. Wah, S. Branson, P. Welinder, P. Perona, and S. Belongie. The Caltech-UCSD Birds-200-2011 Dataset. Technical Re- port CNS-TR-2011-001, California Institute of Technology,

    2011

  31. [40]

    Yu, and Ishan Misra

    Xudong Wang, Rohit Girdhar, Stella X. Yu, and Ishan Misra. Cut and learn for unsupervised object detection and instance segmentation. In CVPR, pages 3124–3134, 2023. 2

    2023

  32. [41]

    Videocutler: Surprisingly simple unsuper- vised video instance segmentation

    Xudong Wang, Ishan Misra, Ziyun Zeng, Rohit Girdhar, and Trevor Darrell. Videocutler: Surprisingly simple unsuper- vised video instance segmentation. In CVPR, pages 22755– 22764, 2024. 2

    2024

  33. [42]

    Crowley, and Dominique Vaufreydaz

    Yangtao Wang, Xi Shen, Shell Xu Hu, Yuan Yuan, James L. Crowley, and Dominique Vaufreydaz. Self-supervised trans- formers for unsupervised object discovery using normalized cut. In CVPR, pages 14523–14533, 2022. 2

    2022

  34. [43]

    Unsupervised part dis- covery via dual representation alignment

    Jiahao Xia, Wenjian Huang, Min Xu, Jianguo Zhang, Haimin Zhang, Ziyu Sheng, and Dong Xu. Unsupervised part dis- covery via dual representation alignment. IEEE TPAMI, pages 1–18, 2024. 1, 2, 3, 6, 7, 8, 11, 13

    2024

  35. [44]

    Hp-capsule: Unsupervised face part discovery by hierarchical parsing capsule network

    Chang Yu, Xiangyu Zhu, Xiaomei Zhang, Zidu Wang, Zhaoxiang Zhang, and Zhen Lei. Hp-capsule: Unsupervised face part discovery by hierarchical parsing capsule network. In CVPR, pages 4022–4031, 2022. 1, 2, 3

    2022

  36. [45]

    How mask matters: Towards theoretical understandings of masked autoencoders

    Qi Zhang, Yifei Wang, and Yisen Wang. How mask matters: Towards theoretical understandings of masked autoencoders. In NIPS, pages 27127–27139, 2022. 2

    2022

  37. [46]

    Unsupervised discovery of object land- marks as structural representations

    Yuting Zhang, Yijie Guo, Yixin Jin, Yijun Luo, Zhiyuan He, and Honglak Lee. Unsupervised discovery of object land- marks as structural representations. In CVPR, pages 2694– 2703, 2018. 1, 3

    2018

  38. [47]

    Pass: Part-aware self-supervised pre- training for person re-identification

    Kuan Zhu, Haiyun Guo, Tianyi Yan, Yousong Zhu, Jinqiao Wang, and Ming Tang. Pass: Part-aware self-supervised pre- training for person re-identification. In ECCV, pages 198– 214, 2022. 2

    2022

  39. [48]

    Adrian Ziegler and Yuki M. Asano. Self-supervised learning of object parts for semantic segmentation. In ICCV, pages 14482–14491, Los Alamitos, CA, USA, 2022. 2 10 Appendix A.1 Dataset Details PartImageNet OOD dataset. We use the OOD variant of PartImageNet [16] to validate MP...

    2022

  40. [49]

    Performance comparison of MPAE with different supervised pretrained backbones on the PartImageNet Segmentation dataset in the setting of K = 50

    with [8] SAM [24] CLIP [31] NMI (%) ↑ 55.10 17.16 33.16 ARI (%) ↑ 73.32 55.26 72.90 Table 12. Performance comparison of MPAE with different supervised pretrained backbones on the PartImageNet Segmentation dataset in the setting of K = 50. The number of MPAE encoder and decoder...

Pith tools

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