RECORD
Labeled Data Selection for Category Discovery
Computer Vision and Pattern Recognition · Submitted 7 Jun 2024
abstract compiled · 2026-09-23 · no language-model review
The authors' main claim Our resulting approaches obtains state-of-the-art discovery performance across a range of challenging fine-grained benchmark datasets.
The compiled record
The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.
Open the compiled record
The abstract, sentence by sentence
Category discovery methods aim to find novel categories in unlabeled visual data. At training time, a set of labeled and unlabeled images are provided, where the labels correspond to the categories present in the images. The labeled data provides guidance during training by indicating what types of visual properties and features are relevant for performing discovery in the unlabeled data. As a result, changing the categories present in the labeled set can have a large impact on what is ultimately discovered in the unlabeled set. Despite its importance, the impact of labeled data selection has not been explored in the category discovery literature to date. We show that changing the labeled data can significantly impact discovery performance. Motivated by this, we propose two new approaches for automatically selecting the most suitable labeled data based on the similarity between the labeled and unlabeled data. Our observation is that, unlike in conventional supervised transfer learning, the best labeled is neither too similar, nor too dissimilar, to the unlabeled categories. Our resulting approaches obtains state-of-the-art discovery performance across a range of challenging fine-grained benchmark datasets.
background contribution method result interpretation main claim
| # | sentence | role | the authors' own work | states the main conclusion | strength |
|---|---|---|---|---|---|
| s0 | Category discovery methods aim to find novel categories in unlabeled visual data. | background 1.00 | 0.04 | 0.02 | asserted 1.00 |
| s1 | At training time, a set of labeled and unlabeled images are provided, where the labels correspond to the categories present in the images. | background 0.90 | 0.09 | 0.02 | asserted 0.99 |
| s2 | The labeled data provides guidance during training by indicating what types of visual properties and features are relevant for performing discovery in the unlabeled data. | background 0.99 | 0.08 | 0.04 | asserted 0.99 |
| s3 | As a result, changing the categories present in the labeled set can have a large impact on what is ultimately discovered in the unlabeled set. | background 0.71 | 0.15 | 0.10 | asserted 0.70 |
| s4 | Despite its importance, the impact of labeled data selection has not been explored in the category discovery literature to date. | gap 1.00 | 0.08 | 0.03 | asserted 1.00 |
| s5 | We show that changing the labeled data can significantly impact discovery performance. | result 0.99 | 0.94 | 0.41 | asserted 1.00 |
| s6 | Motivated by this, we propose two new approaches for automatically selecting the most suitable labeled data based on the similarity between the labeled and unlabeled data. | contribution 0.85 | 0.84 | 0.09 | asserted 0.99 |
| s7 | Our observation is that, unlike in conventional supervised transfer learning, the best labeled is neither too similar, nor too dissimilar, to the unlabeled categories. | result 0.68 | 0.89 | 0.43 | asserted 1.00 |
| s8 | Our resulting approaches obtains state-of-the-art discovery performance across a range of challenging fine-grained benchmark datasets. | result 0.99 | 0.94 | 0.40 | asserted 1.00 |
What was examined, and what was not
- arXiv listing abstract v2
- full text
- figures and tables
- references and citing sentences
- relations to other papers
- P abstract
- S sections
- T body sentences
- C citations
- R cross-paper relations
The paper
Original abstract
Category discovery methods aim to find novel categories in unlabeled visual data. At training time, a set of labeled and unlabeled images are provided, where the labels correspond to the categories present in the images. The labeled data provides guidance during training by indicating what types of visual properties and features are relevant for performing discovery in the unlabeled data. As a result, changing the categories present in the labeled set can have a large impact on what is ultimately discovered in the unlabeled set. Despite its importance, the impact of labeled data selection has not been explored in the category discovery literature to date. We show that changing the labeled data can significantly impact discovery performance. Motivated by this, we propose two new approaches for automatically selecting the most suitable labeled data based on the similarity between the labeled and unlabeled data. Our observation is that, unlike in conventional supervised transfer learning, the best labeled is neither too similar, nor too dissimilar, to the unlabeled categories. Our resulting approaches obtains state-of-the-art discovery performance across a range of challenging fine-grained benchmark datasets.
Human review
How the literature uses it
1 arXiv paper cite this in 1 sentence
Papers citing it
Cites it as support 1 paper
-
However, as demonstrated by [ref] , not all known class data is useful, and therefore, we believe not all learned concepts are beneficial.
Composing Novel Classes: A Concept-Driven Approach to Generalized Category Discovery · Discussion · 1.00
What it cites
Uses data or values from 1 paper
-
SSB contains data from CUB [ref] , Stanford-Cars [ref] , and FGVC-Aircraft [ref] , and provides the category splits of the data for category discovery with an annotation of `Easy', `Medium', and `Hard' categories that are built on the hierarchy from the dataset labels.
Fine-Grained Visual Classification of Aircraft · Implementation details · 0.94
Discussion
Know something about this paper? Leave a comment, or highlight any passage above to comment on it.