pith. sign in

arxiv: 2509.22769 · v2 · pith:66WAEQ3Onew · submitted 2025-09-26 · 💻 cs.CV

PartCo: Part-Level Correspondence Priors Enhance Category Discovery

Fernando Julio Cendra , Kai Han This is my paper

Pith reviewed 2026-05-22 13:27 UTC · model grok-4.3

classification 💻 cs.CV
keywords Generalized Category DiscoveryPart-Level CorrespondenceUnlabeled DataSemantic StructuresVisual FeaturesCategory DiscoveryComputer Vision
0
0 comments X

The pith

Part-level visual correspondences added to existing methods improve generalized category discovery by capturing finer semantic differences.

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

The paper proposes PartCo as a way to strengthen generalized category discovery by feeding part-level feature matches into current algorithms. Existing approaches mostly use whole-image features and labels from known classes, which can blur distinctions among similar but distinct categories in the unlabeled portion of the data. By adding a prior that aligns specific visual parts across images, the method aims to reveal compositional relationships that global representations overlook. This integration requires no large changes to the underlying discovery pipelines. If successful, it would mean better separation of both familiar and entirely new categories on standard test collections.

Core claim

PartCo incorporates part-level visual feature correspondences as a prior that captures finer-grained semantic structures, allowing existing generalized category discovery methods to bridge semantic labels with part-level visual compositions and achieve higher performance on benchmark datasets.

What carries the argument

The Part-Level Correspondence Prior, which extracts and applies relationships between corresponding parts of different images to refine category boundaries.

If this is right

  • Existing GCD methods reach higher accuracy on multiple benchmark datasets once the part-level prior is added.
  • Category relationships become clearer through the added link between semantic labels and part-level visual compositions.
  • New performance benchmarks are established for generalized category discovery.
  • The prior integrates directly with current pipelines and needs no major architectural changes.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Similar part-matching ideas could be tested in other unsupervised vision tasks that also suffer from coarse global features.
  • Systems that must handle novel objects in real scenes, such as robotics or surveillance, might gain from explicit part composition signals.
  • Automatically learning which parts to match without extra labels remains an open extension worth checking on larger, more varied image collections.

Load-bearing premise

That part-level visual feature correspondences can be reliably computed from the data and will consistently help distinguish categories rather than add noise.

What would settle it

Running PartCo on the same benchmark datasets and observing no consistent accuracy gains or outright drops compared to the unmodified baselines would disprove the central claim.

Figures

Figures reproduced from arXiv: 2509.22769 by Fernando Julio Cendra, Kai Han.

Figure 1
Figure 1. Figure 1: Generalized Category Discovery: Given a labeled subset contains seen classes, the task is to categorize the unlabeled images, which may belong to seen or unseen classes. A growing body of literature in GCD emphasizes the significance of object parts as ef￾fective conduits for transferring knowledge between “seen” and “unseen” categories (Vaze ∗Corresponding author. 1 arXiv:2509.22769v1 [cs.CV] 26 Sep 2025 … view at source ↗
Figure 2
Figure 2. Figure 2 [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of part-level correspondence labels construction: This two-step process begins by applying PCA projections to extract object and detailed features from ViT’s patch tokens using a subset of the dataset. These projections are then applied to the entire dataset to generate part-level correspondence labels. Step 1: PCA projections. We begin by sampling a subset of M labeled images from the dataset Dl … view at source ↗
Figure 4
Figure 4. Figure 4: Visualization of our part-level correspondence labels. For each image, we generate both first- and second-order labels. First-order labels suffice for fine-grained datasets, while second-order labels capture additional detail for generic datasets. In practice, selecting between 1st- and 2nd-order is straightforward: datasets with subtle, intra-class differences indicate fine-grained samples, whereas datase… view at source ↗
Figure 5
Figure 5. Figure 5: (a) PartCo framework: Introduces part-level correspondence labels as a plug-and-play module to enhance GCD methods. (b) Part-level correspondence loss: Depicts how part-level correspondence loss is integrated into the model to learn relationships between parts in ViT’s patch token features. Guiding patch token features. For an input image xi , we first extract its patch token features using the foundation … view at source ↗
Figure 6
Figure 6. Figure 6: Average absolute % gain of PartCo over each baseline on SSB benchmark. On SSB per-dataset results, we observe large absolute % gains across all three datasets: FGVC￾Aircraft (DINOv2: +12.6%, DINOv3: +3.9%), CUB (DINOv2: +9.6%, DINOv3: +2.9%), and Stanford-Cars (DINOv2: +7.4%, DINOv3: +4.6%). On generic per-dataset results, gains are smaller but steady. These trends are consistent across DINOv2 and DINOv3, … view at source ↗
Figure 7
Figure 7. Figure 7: Ablation study investigating the impact of different order levels in part-level correspondence [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Impact of the unsupervised part-level corre [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
read the original abstract

Generalized Category Discovery (GCD) aims to identify both known and novel categories within unlabeled data by leveraging a set of labeled examples from known categories. Existing GCD methods primarily depend on semantic labels and global image representations, often overlooking the detailed part-level cues that are crucial for distinguishing closely related categories. In this paper, we introduce PartCo, short for Part-Level Correspondence Prior, a novel framework that enhances category discovery by incorporating part-level visual feature correspondences. By leveraging part-level relationships, PartCo captures finer-grained semantic structures, enabling a more nuanced understanding of category relationships. Importantly, PartCo seamlessly integrates with existing GCD methods without requiring significant modifications. Our extensive experiments on multiple benchmark datasets demonstrate that PartCo significantly improves the performance of current GCD approaches, outperforming most existing methods by bridging the gap between semantic labels and part-level visual compositions, thereby setting new benchmarks for GCD.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces PartCo, a framework for Generalized Category Discovery (GCD) that augments existing methods with part-level visual feature correspondence priors. It claims these priors capture finer-grained semantic structures, enable better distinction of closely related categories, integrate seamlessly without major modifications to base GCD pipelines, and yield consistent outperformance on standard benchmarks for both known and novel classes.

Significance. If the central claims hold, the work would be moderately significant for the GCD literature by shifting attention from purely global or label-driven representations toward part-level visual composition. The emphasis on plug-and-play integration is a practical strength that could encourage adoption; however, the absence of explicit validation that part correspondences transfer reliably to unlabeled novel categories limits the strength of the contribution.

major comments (2)
  1. [§3] §3 (Method): The description of the part-level correspondence extraction and its integration into the GCD objective does not include an explicit mechanism or ablation demonstrating that the correspondences are derived without implicit leakage from the labeled known-category data. This is load-bearing for the claim that PartCo bridges semantic labels to part-level compositions for novel categories.
  2. [§4.2, Table 2] §4.2, Table 2 (Novel-category results): The reported gains on novel classes are presented without error bars, multiple random seeds, or a statistical significance test against the strongest baseline; the absolute improvements are modest and could be consistent with run-to-run variance, weakening the assertion of setting new benchmarks.
minor comments (2)
  1. [Figure 3] Figure 3: The visualization of part correspondences would benefit from clearer annotation of which parts are matched across images and whether the matches are for known or novel categories.
  2. [§3.1] Notation in §3.1: The symbol for the correspondence prior is introduced without an explicit definition of its dimensionality or normalization, making the subsequent equations harder to follow.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback on our manuscript. We address each major comment below with point-by-point responses and indicate the revisions we will incorporate to strengthen the paper.

read point-by-point responses

  1. Referee: [§3] §3 (Method): The description of the part-level correspondence extraction and its integration into the GCD objective does not include an explicit mechanism or ablation demonstrating that the correspondences are derived without implicit leakage from the labeled known-category data. This is load-bearing for the claim that PartCo bridges semantic labels to part-level compositions for novel categories.

    Authors: We appreciate the referee drawing attention to this critical aspect of our method. The part-level correspondences in PartCo are computed using a frozen, pre-trained part-aware feature extractor (based on self-supervised visual representations) that processes raw image patches without any access to or supervision from the semantic labels of the known categories. The extractor is applied uniformly to both labeled and unlabeled data in a label-agnostic manner, ensuring that the priors capture intrinsic visual part structures rather than label-derived information. To make this explicit and directly address potential concerns about leakage or transfer to novel categories, we will revise §3 to include a dedicated subsection detailing the extraction pipeline, its independence from labels, and an ablation study that isolates the contribution of label-free versus label-informed priors. This revision will further substantiate the claim that PartCo enables effective bridging to novel categories. revision: yes

  2. Referee: [§4.2, Table 2] §4.2, Table 2 (Novel-category results): The reported gains on novel classes are presented without error bars, multiple random seeds, or a statistical significance test against the strongest baseline; the absolute improvements are modest and could be consistent with run-to-run variance, weakening the assertion of setting new benchmarks.

    Authors: We acknowledge the value of reporting statistical robustness for the novel-class results. Although the improvements in Table 2 are consistent across multiple datasets and align with the overall trend of better fine-grained discrimination, we agree that the absence of variability measures limits the strength of the claims. In the revised manuscript, we will rerun the key experiments using at least three random seeds, report means and standard deviations in Table 2 (and related tables), and include paired statistical significance tests (e.g., t-tests) against the strongest baselines. These additions will provide clearer evidence that the observed gains exceed run-to-run variance while preserving the original experimental setup. revision: yes

Circularity Check

0 steps flagged

No significant circularity; framework augments existing GCD methods with independent part-level priors

full rationale

The paper introduces PartCo as a modular addition to existing Generalized Category Discovery approaches, leveraging part-level visual feature correspondences to capture finer semantic structures. No derivation chain, equations, or predictions are presented that reduce by construction to fitted inputs or self-referential definitions. The central claims rest on empirical integration and benchmark improvements rather than any self-definitional loop, uniqueness theorem from prior self-work, or ansatz smuggled via citation. The framework is described as seamlessly integrable without significant modifications, indicating the contribution is additive and externally testable rather than tautological.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that part-level correspondences supply semantic information beyond global image features.

axioms (1)
  • domain assumption Part-level visual feature correspondences can be extracted and used to capture finer-grained semantic structures for category discovery.
    Invoked as the core mechanism enabling nuanced category relationships.

pith-pipeline@v0.9.0 · 5675 in / 1036 out tokens · 40025 ms · 2026-05-22T13:27:46.583364+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Generalized Category Discovery under Domain Shifts: From Vision to Vision-Language Models

    cs.CV 2026-04 unverdicted novelty 6.0

    Three frameworks adapt foundation models for generalized category discovery under domain shifts via disentanglement and prompt tuning, showing gains on synthetic and real multi-domain data.

Reference graph

Works this paper leans on

64 extracted references · 64 canonical work pages · cited by 1 Pith paper · 2 internal anchors

  1. [1]

    write newline

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

    work page

  2. [2]

    Deep vit features as dense visual descriptors

    Shir Amir, Yossi Gandelsman, Shai Bagon, and Tali Dekel. Deep vit features as dense visual descriptors. In ECCV workshop, 2022

    work page 2022

  3. [3]

    Semi-supervised learning of visual features by non-parametrically predicting view assignments with support samples

    Mahmoud Assran, Mathilde Caron, Ishan Misra, Piotr Bojanowski, Armand Joulin, Nicolas Ballas, and Michael Rabbat. Semi-supervised learning of visual features by non-parametrically predicting view assignments with support samples. In ICCV, 2021

    work page 2021

  4. [4]

    Amend: Adaptive margin and expanded neighborhood for efficient generalized category discovery

    Anwesha Banerjee, Liyana Sahir Kallooriyakath, and Soma Biswas. Amend: Adaptive margin and expanded neighborhood for efficient generalized category discovery. In WACV, 2024

    work page 2024

  5. [5]

    Mixmatch: A holistic approach to semi-supervised learning

    David Berthelot, Nicholas Carlini, Ian Goodfellow, Nicolas Papernot, Avital Oliver, and Colin Raffel. Mixmatch: A holistic approach to semi-supervised learning. In NeurIPS, 2019

    work page 2019

  6. [6]

    Bird species categorization using pose normalized deep convolutional nets

    Steve Branson, Grant Van Horn, Serge Belongie, and Pietro Perona. Bird species categorization using pose normalized deep convolutional nets. In BMVC, 2014

    work page 2014

  7. [7]

    Open-world semi-supervised learning

    Kaidi Cao, Maria Brbic, and Jure Leskovec. Open-world semi-supervised learning. In ICLR, 2022

    work page 2022

  8. [8]

    Emerging properties in self-supervised vision transformers

    Mathilde Caron, Hugo Touvron, Ishan Misra, Herve Jegou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerging properties in self-supervised vision transformers. In ICCV, 2021

    work page 2021

  9. [9]

    Promptccd: Learning gaussian mixture prompt pool for continual category discovery

    Fernando Julio Cendra, Bingchen Zhao, and Kai Han. Promptccd: Learning gaussian mixture prompt pool for continual category discovery. In ECCV, 2024

    work page 2024

  10. [10]

    Semi-supervised learning (chapelle, o

    Olivier Chapelle, Bernhard Scholkopf, and Alexander Zien. Semi-supervised learning (chapelle, o. et al., eds.; 2006)[book reviews]. IEEE Transactions on Neural Networks, 2009

    work page 2006

  11. [11]

    Fine-grained image-text correspondence with cost aggregation for open-vocabulary part segmentation

    Jiho Choi, Seonho Lee, Minhyun Lee, Seungho Lee, and Hyunjung Shim. Fine-grained image-text correspondence with cost aggregation for open-vocabulary part segmentation. In CVPR, 2025

    work page 2025

  12. [12]

    Unsupervised part discovery from contrastive reconstruction

    Subhabrata Choudhury, Iro Laina, Christian Rupprecht, and Andrea Vedaldi. Unsupervised part discovery from contrastive reconstruction. In NeurIPS, 2021

    work page 2021

  13. [13]

    Imagenet: A large-scale hierarchical image database

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

    work page 2009

  14. [14]

    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, Sylvain Gelly, et al. An image is worth 16x16 words: Transformers for image recognition at scale. In ICLR, 2021

    work page 2021

  15. [15]

    A unified objective for novel class discovery

    Enrico Fini, Enver Sangineto, St \'e phane Lathuiliere, Zhun Zhong, Moin Nabi, and Elisa Ricci. A unified objective for novel class discovery. In ICCV, 2021

    work page 2021

  16. [16]

    Learning to discover novel visual categories via deep transfer clustering

    Kai Han, Andrea Vedaldi, and Andrew Zisserman. Learning to discover novel visual categories via deep transfer clustering. In ICCV, 2019

    work page 2019

  17. [17]

    Automatically discovering and learning new visual categories with ranking statistics

    Kai Han, Sylvestre-Alvise Rebuffi, Sebastien Ehrhardt, Andrea Vedaldi, and Andrew Zisserman. Automatically discovering and learning new visual categories with ranking statistics. In ICLR, 2020

    work page 2020

  18. [18]

    Autonovel: Automatically discovering and learning novel visual categories

    Kai Han, Sylvestre-Alvise Rebuffi, Sebastien Ehrhardt, Andrea Vedaldi, and Andrew Zisserman. Autonovel: Automatically discovering and learning novel visual categories. IEEE TPAMI, 2021

    work page 2021

  19. [19]

    Cipr: An efficient framework with cross-instance positive relations for generalized category discovery

    Shaozhe Hao, Kai Han, and Kwan-Yee K Wong. Cipr: An efficient framework with cross-instance positive relations for generalized category discovery. TMLR, 2024

    work page 2024

  20. [20]

    Trash to treasure: harvesting ood data with cross-modal matching for open-set semi-supervised learning

    Junkai Huang, Chaowei Fang, Weikai Chen, Zhenhua Chai, Xiaolin Wei, Pengxu Wei, Liang Lin, and Guanbin Li. Trash to treasure: harvesting ood data with cross-modal matching for open-set semi-supervised learning. In ICCV, 2021

    work page 2021

  21. [21]

    Joint representation learning and novel category discovery on single-and multi-modal data

    Xuhui Jia, Kai Han, Yukun Zhu, and Bradley Green. Joint representation learning and novel category discovery on single-and multi-modal data. In ICCV, 2021

    work page 2021

  22. [22]

    Novel class discovery without forgetting

    KJ Joseph, Sujoy Paul, Gaurav Aggarwal, Soma Biswas, Piyush Rai, Kai Han, and Vineeth N Balasubramanian. Novel class discovery without forgetting. In ECCV, 2022

    work page 2022

  23. [23]

    3d object representations for fine-grained categorization

    Jonathan Krause, Michael Stark, Jia Deng, and Li Fei-Fei. 3d object representations for fine-grained categorization. In ICCV workshop, 2013

    work page 2013

  24. [24]

    Krizhevsky and G

    A. Krizhevsky and G. Hinton. Learning multiple layers of features from tiny images. Master's thesis, Department of Computer Science, University of Toronto, 2009

    work page 2009

  25. [25]

    Temporal ensembling for semi-supervised learning

    Samuli Laine and Timo Aila. Temporal ensembling for semi-supervised learning. In ICLR, 2017

    work page 2017

  26. [26]

    Flipped classroom: Aligning teacher attention with student in generalized category discovery

    Haonan Lin, Wenbin An, Jiahao Wang, Yan Chen, Feng Tian, Mengmeng Wang, QianYing Wang, Guang Dai, and Jingdong Wang. Flipped classroom: Aligning teacher attention with student in generalized category discovery. In NeurIPS, 2024

    work page 2024

  27. [27]

    Debgcd: Debiased learning with distribution guidance for generalized category discovery

    Yuanpei Liu and Kai Han. Debgcd: Debiased learning with distribution guidance for generalized category discovery. In ICLR, 2025

    work page 2025

  28. [28]

    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

    work page 2019

  29. [29]

    Protogcd: Unified and unbiased prototype learning for generalized category discovery

    Shijie Ma, Fei Zhu, Xu-Yao Zhang, and Cheng-Lin Liu. Protogcd: Unified and unbiased prototype learning for generalized category discovery. IEEE TPAMI, 2025

    work page 2025

  30. [30]

    Fine-Grained Visual Classification of Aircraft

    Subhransu Maji, Esa Rahtu, Juho Kannala, Matthew Blaschko, and Andrea Vedaldi. Fine-grained visual classification of aircraft. arXiv preprint arXiv:1306.5151, 2013

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  31. [31]

    Discovering relationships between object categories via universal canonical maps

    Natalia Neverova, Artsiom Sanakoyeu, Patrick Labatut, David Novotny, and Andrea Vedaldi. Discovering relationships between object categories via universal canonical maps. In CVPR, 2021

    work page 2021

  32. [32]

    I have seen enough: Transferring parts across categories

    David Novotny, Diane Larlus, and Andrea Vedaldi. I have seen enough: Transferring parts across categories. In BMVC, 2016

    work page 2016

  33. [33]

    AnchorNet : A weakly supervised network to learn geometry-sensitive features for semantic matching

    David Novotny, Diane Larlus, and Andrea Vedaldi. AnchorNet : A weakly supervised network to learn geometry-sensitive features for semantic matching. In CVPR, 2017

    work page 2017

  34. [34]

    Self-supervised learning of geometrically stable features through probabilistic introspection

    David Novotny, Samuel Albanie, Diane Larlus, and Andrea Vedaldi. Self-supervised learning of geometrically stable features through probabilistic introspection. In CVPR, 2018

    work page 2018

  35. [35]

    Realistic evaluation of deep semi-supervised learning algorithms

    Avital Oliver, Augustus Odena, Colin A Raffel, Ekin Dogus Cubuk, and Ian Goodfellow. Realistic evaluation of deep semi-supervised learning algorithms. In NeurIPS, 2018

    work page 2018

  36. [36]

    Maxime Oquab, Timoth \'e e Darcet, Th \'e o Moutakanni, Huy V. Vo, 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 Rabbat, Vasu Sharma, Gabriel Synnaeve, Hu Xu, Herve Jegou, Julien Mairal, Pat...

    work page 2024

  37. [37]

    Cats and dogs

    Omkar M Parkhi, Andrea Vedaldi, Andrew Zisserman, and CV Jawahar. Cats and dogs. In CVPR, 2012

    work page 2012

  38. [38]

    Dynamic conceptional contrastive learning for generalized category discovery

    Nan Pu, Zhun Zhong, and Nicu Sebe. Dynamic conceptional contrastive learning for generalized category discovery. In CVPR, 2023

    work page 2023

  39. [39]

    Federated generalized category discovery

    Nan Pu, Wenjing Li, Xingyuan Ji, Yalan Qin, Nicu Sebe, and Zhun Zhong. Federated generalized category discovery. In CVPR, 2024

    work page 2024

  40. [40]

    Selex: Self-expertise in fine-grained generalized category discovery

    Sarah Rastegar, Mohammadreza Salehi, Yuki M Asano, Hazel Doughty, and Cees G M Snoek. Selex: Self-expertise in fine-grained generalized category discovery. In ECCV, 2024

    work page 2024

  41. [41]

    In defense of pseudo-labeling: An uncertainty-aware pseudo-label selection framework for semi-supervised learning

    Mamshad Nayeem Rizve, Kevin Duarte, Yogesh S Rawat, and Mubarak Shah. In defense of pseudo-labeling: An uncertainty-aware pseudo-label selection framework for semi-supervised learning. In ICLR, 2021

    work page 2021

  42. [42]

    Openmatch: Open-set consistency regularization for semi-supervised learning with outliers

    Kuniaki Saito, Donghyun Kim, and Kate Saenko. Openmatch: Open-set consistency regularization for semi-supervised learning with outliers. In NeurIPS, 2021

    work page 2021

  43. [43]

    Oriane Sim \'e oni, Huy V Vo, Maximilian Seitzer, Federico Baldassarre, Maxime Oquab, Cijo Jose, Vasil Khalidov, Marc Szafraniec, Seungeun Yi, Micha \"e l Ramamonjisoa, et al. Dinov3. arXiv preprint arXiv:2508.10104, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  44. [44]

    Fixmatch: Simplifying semi-supervised learning with consistency and confidence

    Kihyuk Sohn, David Berthelot, Nicholas Carlini, Zizhao Zhang, Han Zhang, Colin A Raffel, Ekin Dogus Cubuk, Alexey Kurakin, and Chun-Liang Li. Fixmatch: Simplifying semi-supervised learning with consistency and confidence. In NeurIPS, 2020

    work page 2020

  45. [45]

    Going denser with open-vocabulary part segmentation

    Peize Sun, Shoufa Chen, Chenchen Zhu, Fanyi Xiao, Ping Luo, Saining Xie, and Zhicheng Yan. Going denser with open-vocabulary part segmentation. In ICCV, 2023

    work page 2023

  46. [46]

    The herbarium challenge 2019 dataset

    Kiat Chuan Tan, Yulong Liu, Barbara Ambrose, Melissa Tulig, and Serge Belongie. The herbarium challenge 2019 dataset. In CVPR workshop, 2019

    work page 2019

  47. [47]

    Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results

    Antti Tarvainen and Harri Valpola. Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results. In NeurIPS, 2017

    work page 2017

  48. [48]

    Generalized category discovery

    Sagar Vaze, Kai Han, Andrea Vedaldi, and Andrew Zisserman. Generalized category discovery. In CVPR, 2022 a

    work page 2022

  49. [49]

    The semantic shift benchmark

    Sagar Vaze, Kai Han, Andrea Vedaldi, and Andrew Zisserman. The semantic shift benchmark. In ICML workshop, 2022 b

    work page 2022

  50. [50]

    No representation rules them all in category discovery

    Sagar Vaze, Andrea Vedaldi, and Andrew Zisserman. No representation rules them all in category discovery. In NeurIPS, 2023

    work page 2023

  51. [51]

    The caltech-ucsd birds-200-2011 dataset

    Catherine Wah, Steve Branson, Peter Welinder, Pietro Perona, and Serge Belongie. The caltech-ucsd birds-200-2011 dataset. California Institute of Technology, 2011

    work page 2011

  52. [52]

    Sptnet: An efficient alternative framework for generalized category discovery with spatial prompt tuning

    Hongjun Wang, Sagar Vaze, and Kai Han. Sptnet: An efficient alternative framework for generalized category discovery with spatial prompt tuning. In ICLR, 2024

    work page 2024

  53. [53]

    Hilo: A learning framework for generalized category discovery robust to domain shifts

    Hongjun Wang, Sagar Vaze, and Kai Han. Hilo: A learning framework for generalized category discovery robust to domain shifts. In ICLR, 2025

    work page 2025

  54. [54]

    Parametric classification for generalized category discovery: A baseline study

    Xin Wen, Bingchen Zhao, and Xiaojuan Qi. Parametric classification for generalized category discovery: A baseline study. In ICCV, 2023

    work page 2023

  55. [55]

    Multi-task curriculum framework for open-set semi-supervised learning

    Qing Yu, Daiki Ikami, Go Irie, and Kiyoharu Aizawa. Multi-task curriculum framework for open-set semi-supervised learning. In ECCV, 2020

    work page 2020

  56. [56]

    A tale of two features: Stable diffusion complements dino for zero-shot semantic correspondence

    Junyi Zhang, Charles Herrmann, Junhwa Hur, Luisa Polania Cabrera, Varun Jampani, Deqing Sun, and Ming-Hsuan Yang. A tale of two features: Stable diffusion complements dino for zero-shot semantic correspondence. In NeurIPS, 2023

    work page 2023

  57. [57]

    Grow and merge: a unified framework for continuous categories discovery

    Xinwei Zhang, Jianwen Jiang, Yutong Feng, Zhi-fan Wu, Xibin Zhao, Hai Wan, Mingqian Tang, Rong Jin, and Yue Gao. Grow and merge: a unified framework for continuous categories discovery. In NeurIPS, 2022

    work page 2022

  58. [58]

    Novel visual category discovery with dual ranking statistics and mutual knowledge distillation

    Bingchen Zhao and Kai Han. Novel visual category discovery with dual ranking statistics and mutual knowledge distillation. In NeurIPS, 2021

    work page 2021

  59. [59]

    Learning semi-supervised gaussian mixture models for generalized category discovery

    Bingchen Zhao, Xin Wen, and Kai Han. Learning semi-supervised gaussian mixture models for generalized category discovery. In ICCV, 2023

    work page 2023

  60. [60]

    Neighborhood contrastive learning for novel class discovery

    Zhun Zhong, Enrico Fini, Subhankar Roy, Zhiming Luo, Elisa Ricci, and Nicu Sebe. Neighborhood contrastive learning for novel class discovery. In CVPR, 2021

    work page 2021

  61. [61]

    Semi-supervised learning literature survey

    Xiaojin Jerry Zhu. Semi-supervised learning literature survey. University of Wisconsin-Madison Department of Computer Sciences, 2005

    work page 2005

  62. [62]

    @esa (Ref

    \@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should not add it explicitly Type <Return> for now, but then later remove the command n...

    work page

  63. [63]

    \@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@first@sw \@firstoftwo \@ifundefined NAT@b*@#2 \@firstoftwo @num @NAT@ctr \@secondoft...

    work page

  64. [64]

    @open @close @open @close and [1] URL: #1 \@ifundefined chapter * \@mkboth \@ifxundefined @sectionbib * \@mkboth * \@mkboth\@gobbletwo \@ifclassloaded amsart * \@ifclassloaded amsbook * \@ifxundefined @heading @heading NAT@ctr thebibliography [1] @ \@biblabel @NAT@ctr \@bibsetup #1 @NAT@ctr @ @openbib .11em \@plus.33em \@minus.07em 4000 4000 `\.\@m @bibit...

    work page arXiv 2011