Pith. sign in

REVIEW 4 major objections 4 minor 67 references

Single-Pass Object-Focused Data Selection

T0 review · 4 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read By clustering foundation-model object features per class and choosing one object per cluster, Object-Focused Data Selection beats random selection and eight baselines for detection and segmentation labeling budgets.

desk verdict OFDS is a genuine object-level data selection method with strong imbalanced-class results, but the abstract overstates consistency on balanced settings and the paper never measures the object proposer's recall, which the coverage story depends on. read the letter →

arxiv 2412.10032 v2 pith:XUQXPVRN submitted 2024-12-13 cs.CV

Niclas Popp , Dan Zhang , Jan Hendrik Metzen , Matthias Hein , Lukas Schott This is my paper · ORCID
classification cs.CV
keywords dataselectionobjectdetectionsemanticsegmentationfoundationmodelsclassimbalanceannotationbudgetactivelearningk-meansclustering
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 asks which unlabeled images to send to human annotators when building a training set for object detection or segmentation under a fixed annotation budget. It claims that a single-pass selection method operating on object-level features, rather than whole-image features, can reliably beat random selection and eight existing baselines. The key move is to cluster foundation-model object features per target class and pick one object from each cluster, so that the selected images cover the semantic variety of every class. This matters because annotation budgets are costly and real-world data is class-imbalanced; the paper reports consistent gains, with the largest gains on rare classes and on several datasets. It also shows that combining these human labels with foundation-model autolabels gives the best results under tight budgets.

What carries the argument

The load-bearing mechanism is class-wise adaptive k-means clustering of foundation-model object features. Each cluster represents a semantically coherent group of objects within a target class; selecting the object nearest to each cluster center yields a density-based covering of that class's feature space, while processing classes by ascending proposal count protects rare classes. The adaptive k (Eq. 2) converts the remaining annotation budget into a number of clusters, and the search for clusters without already-annotated images prevents near-duplicates.

What would settle it

Evaluate OFDS on a dataset where a target class is present but reliably missed by the object proposer (e.g., a specialized remote-sensing class with low recall at the 5% false-positive threshold). If training on OFDS-selected subsets does not beat random selection for that class, it confirms the proposed selection inherits proposer recall failures. A sharper test: replace Grounding DINO proposals with oracle human-annotated boxes as the proposer input; if downstream performance does not improve over the standard proposer, then the bottleneck is elsewhere (e.g., clustering or features), not proposal recall.

Watch

Extended reading notes

Core claim

OFDS selects images to label by first running an open-world detector (Grounding DINO) to propose objects, extracting 256-dimensional object features from SAM 2, and assigning each proposal to a target class by its predicted label. For each class it clusters those features with k-means, choosing k adaptively from the remaining annotation budget and the number of already selected images, and then selects one object nearest to each cluster centroid. The images containing those selected objects are exhaustively labeled by humans. Because classes are processed in ascending order of proposal count, rare classes get covered first. The paper reports that this object-level semantic covering consistently outperforms random selection and all baselines across object detection, semantic segmentation, and instance segmentation on PASCAL VOC (balanced and artificially imbalanced), Cityscapes, and LoveDA, and that the margin is largest for rare classes and for class-imbalanced settings.

Load-bearing premise

The whole selection rests on the object proposer's proposals and the SAM2 features accurately representing the true objects of every target class; if the proposer misses a class or an object appearance, that class or appearance is never clustered and never selected, because OFDS only controls the false-positive rate, not the false-negative rate.

Editorial extensions

If this is right

  • Under a fixed annotation budget measured in annotation units, training a compact detector or segmenter on an OFDS-selected subset yields higher mAP or mIoU than training on subsets selected by random drawing or by any of the eight image-level baselines tested.
  • The advantage over baselines is largest for the rarest classes, meaning OFDS mitigates class imbalance at the data-selection stage rather than through training-time loss adjustments.
  • Pre-training a downstream model on full-dataset autolabels from foundation models and then fine-tuning on an OFDS-selected human-labeled subset outperforms either autolabels alone or human labels alone for the same annotation cost.
  • Using OFDS to choose the initial labeled set for active learning improves the entire active learning curve compared with a random cold start or HaCON.
  • OFDS is deterministic and requires no labeled data and no downstream model training during selection; selection is a single pass with cost comparable to one forward inference pass.

Reading between the lines

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

  • Because OFDS relies entirely on the object proposer's recall, its gains should scale with foundation-model improvements; a better proposer should translate directly into better selection, and the paper's own ablation with a larger backbone supports this extrapolation.
  • The method suggests a testable recipe for other dense prediction tasks: any task whose annotation unit is an object-like region (e.g., panoptic parts, medical lesions) could use the same cluster-and-cover selection, provided a proposer and feature extractor exist for that domain.
  • The near-duplicate-removal property of cluster-center selection may be valuable beyond annotation budgeting, e.g., in cleaning web-scale image collections before self-supervised pre-training, though the paper only demonstrates the duplication-robustness in a synthetic duplicate setting.
  • The authors do not claim OFDS replaces active learning; their direct comparison shows it is on par with the tested active learning frameworks while being model-agnostic, which suggests data selection and active learning are complementary rather than competing stages.
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

4 major / 4 minor

Summary. The paper introduces Object-Focused Data Selection (OFDS), a single-pass, training-free method for selecting images to label for object detection and segmentation under a budget measured in annotation units. OFDS uses Grounding DINO to propose objects, SAM 2 to extract object-level features, per-class adaptive k-means clustering, and selects one object per cluster so that the chosen images semantically cover each target class. The method is evaluated on PASCAL VOC (including a deliberately class-imbalanced variant), Cityscapes, and LoveDA, against eight single-pass baselines, across multiple downstream architectures, and as a warm start for active learning. The authors report that OFDS consistently outperforms random selection and baselines, especially under class imbalance, and that combining OFDS-selected human labels with foundation-model autolabels gives the best results at constrained budgets.

Significance. If the empirical claims hold, OFDS addresses a practical and under-studied problem: selecting annotation units for dense prediction tasks before any downstream model is trained. The paper has several genuine strengths: it moves data selection from image-level to object-level features, explicitly models annotation-unit budgets, evaluates a broad set of baselines across multiple tasks and datasets, includes ablations on the clustering step, object proposer strength, downstream architectures, and repeated random/FreeSel runs for one dataset, and provides class-balance analysis of the selected subsets. The object-level coverage idea is intuitive and the imbalanced-setting results are consistently favorable. However, the central claim of universal superiority over random selection is not fully supported by the paper's own balanced-setting tables, and the mechanism's reliance on object-proposer recall is not tested, which is a load-bearing risk for the rare-class and specialized-domain scenarios the paper emphasizes.

major comments (4)
  1. [Abstract; Sec. 4.2; Table 7 in App. Q] The abstract states that OFDS "consistently outperforms random selection and all baselines," but the paper's own balanced-setting results do not support this. In Table 7 (semantic segmentation on the original balanced PASCAL VOC split), OFDS has mIoU 48.52 versus random 49.60 at the 10% budget and 61.89 versus 61.93 at the 40% budget; at several other budgets the differences are small and not consistently in OFDS's favor. The main-text discussion in Sec. 4.2 correctly acknowledges that OFDS performs "on par" with random on full PASCAL VOC. Please revise the abstract and conclusion to state the qualified version of the claim.
  2. [Algorithm 1, Step 1; Sec. 3.2; App. I.2; Limitations] The coverage mechanism presumes that the object proposer has high recall for every target class, but recall is never measured. The confidence threshold is calibrated to a 5% false-positive rate (App. I.2), which deliberately trades recall for precision, and the Limitations section concedes that OFDS "inherits [the object proposer's] biases and limitations." For any class that the proposer systematically misses, the per-class feature set in Step 5 of Algorithm 1 is empty or incomplete, so the selected subset can omit that class entirely, while random selection would still include some of its images. This is precisely the risk for the rare-class and specialized-domain settings the paper targets. Please report per-class recall of the object proposer on the target validation sets and, if necessary, adjust the selection procedure or threshold so that all target classes are verifiably present in the selected subset.
  3. [Eq. (2); Algorithm 1 input] The budget update in Eq. (2) depends on N_O, the estimated average number of target objects per image, but the main text does not state how N_O is obtained or validated. The reported actual subset sizes are close to the target sizes, which suggests N_O is set from data statistics, but no estimation procedure or sensitivity analysis is provided. Since an inaccurate N_O directly changes the number of images selected and the per-class budget distribution, please document how N_O is computed for each dataset and include an ablation or robustness check.
  4. [App. J (Implementation Details)] The implementation paragraph says that "we only consider object proposals with bounding boxes smaller than 0.05% of the overall image area to filter out noisy proposals." Interpreted literally, this removes small proposals, which are often exactly the rare and difficult objects the method aims to cover. If the intended condition is that proposals larger than this threshold are removed, please correct the text; otherwise, justify why filtering out small proposals is consistent with the claimed semantic coverage of all target classes.
minor comments (4)
  1. [Sec. 4.1, Baselines paragraph] The sentence "they only work in the image feature space" contains a typo and should read "they only work in the image feature space."
  2. [Sec. 3.2; App. H.1] The term "annotation units" is used for both bounding boxes and masks, but for semantic segmentation multiple masks on the same image can count separately, as noted in App. H.1. Please define this counting convention consistently when the annotation budget is introduced in Sec. 3.
  3. [App. J, adaptive cluster search] The adaptive cluster search multiplies the number of clusters by 1.05 until enough clusters are found, but the choice of the factor is not motivated or ablated. A brief discussion of its influence would help.
  4. [App. F] Repeated selection runs are reported only for PASCAL VOC with class imbalance. Since random selection and FreeSel are stochastic, it would be helpful to state that single runs are used elsewhere or to provide repeated-run intervals for at least one additional dataset.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: OFDS is unsupervised and externally calibrated, so its downstream gains are not equivalent to its inputs.

full rationale

The claimed derivation chain is empirical rather than definitional. OFDS's only tunable constants come from external calibration: the object-proposer confidence threshold is set to a 5% false-positive rate on an MSCOCO reference set (Sec. 3.2, App. I), not on target labels or downstream metrics. The clustering (k-means on SAM2 object features) is unsupervised, and the per-class budget (Eq. 2) is computed from the annotation budget B and an estimate NO; no parameter is fitted to downstream mAP or mIoU. The selection does not optimize or regress the reported performance numbers, so the improved rare-class performance is a consequence of per-class balanced inclusion rather than a renamed fit. The same foundation-model family is used for autolabel pre-training and OFDS features in Sec. 4.3, which is a shared-bias channel that could affect the combined result, but this is not a reduction of the result to its inputs because OFDS never uses autolabels or the downstream model during selection. The paper's stated limitation—'OFDS depends on features generated by the object proposer and thereby inherits its biases and limitations' (Limitations section)—acknowledges the proposer-recall assumption; low recall would weaken the coverage mechanism, but this is a validity/robustness concern, not circularity. No uniqueness theorem or load-bearing self-citation is invoked, and no equation is shown to be equivalent to an input by construction.

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

The method rests on foundation-model proposal quality, feature geometry, and calibration transfer. It adds no physical or mathematical invented entities. The main uncharged inputs are the calibrated threshold, the undisclosed N_O estimate, and the assumption that k-means on SAM2 features gives semantic coverage.

free parameters (4)
  • Object proposer confidence threshold = not stated; calibrated to 5% false positive rate on MSCOCO reference classes
    Controls precision and recall of Grounding DINO proposals used for all clustering and selection; the threshold is chosen via a reference set rather than derived (Section 3.2, App. I).
  • Estimated annotation units per image N_O = not stated
    Eq. 2 divides the per-class budget by N_O to get the number of images; incorrect values make the selected subset miss the target budget. The paper does not describe how N_O is estimated or set.
  • Adaptive cluster search factor = 1.05
    App. J iteratively multiplies the number of clusters by 1.05 until enough clusters without annotated objects exist; a heuristic not derived from data.
  • Bounding box area filter = 0.05% of image area
    App. J discards proposals smaller than 0.05% of image area to filter noise, changing which objects enter clustering and selection.
assumptions (4)
  • domain assumption SAM 2 object features are semantically coherent under Euclidean k-means clustering.
    Section 3.2 and Fig. 12 use cluster centroids to represent semantic groups; there is no proof that cluster coverage equals semantic coverage.
  • domain assumption Grounding DINO proposals have sufficient recall after fixing a 5% false positive rate.
    Step 1 of Algorithm 1 discards proposals below the calibrated threshold; missed objects never enter selection, so recall is not controlled and is implicitly assumed adequate.
  • domain assumption Class names provided by the user map reliably to the object proposer's text embeddings across domains.
    Used in Step 1; LoveDA's remote-sensing classes are acknowledged as out-of-distribution and the gains shrink, which is evidence that the assumption degrades.
  • domain assumption FPR calibration on manually selected MSCOCO classes transfers to the target datasets.
    App. I.1 manually maps target classes to MSCOCO; for LoveDA the Cityscapes thresholds are reused because no calibration data exists.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Single-Pass Object-Focused Data Selection." pith.science (2026). https://pith.science/paper/XUQXPVRN

@misc{pith2026241210032,
  author       = {Pith},
  title        = {Pith review of: Single-Pass Object-Focused Data Selection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XUQXPVRN}},
  note         = {Machine review of arXiv:2412.10032}
}
read the original abstract

While unlabeled image data is often plentiful, the costs of high-quality labels pose an important practical challenge: Which images should one select for labeling to use the annotation budget for a particular target task most effectively? To address this problem, we focus on single-pass data selection, which refers to the process of selecting all data to be annotated at once before training a downstream model. Prior methods for single-pass data selection rely on image-level representations and fail to reliably outperform random selection for object detection and segmentation. We propose Object-Focused Data Selection (OFDS) which leverages object-level features from foundation models and ensures semantic coverage of all target classes. In extensive experiments across tasks and target domains, OFDS consistently outperforms random selection and all baselines. The best results for constrained annotation budgets are obtained by combining human labels from OFDS with autolabels from foundation models. Moreover, using OFDS to select the initial labeled set for active learning yields consistent improvements

Figures

Figures reproduced from arXiv: 2412.10032 by the authors.

Figure 1
Figure 1. Setup for Object-Focused Data Selection. OFDS selects a subset of images to be labeled from an unlabeled dataset given a set of object classes and an annotation budget for object-level annotations. Training downstream models with the data selected by OFDS improves over all data selection baselines and random selection (results are for an annotation budget of 5% of all annotation units for the PASCAL VOC dataset with… view at source ↗
Figure 2
Figure 2. Two Stage Selection Process of OFDS. In the first step, coarse object-level features are extracted from foundation models. These features are grouped per class and clustered. The number of clusters is chosen adaptively based on the remaining budget and already labeled objects. For every cluster, we select one object to ensure a semantic covering of the target classes. Only images containing selected objects are labe… view at source ↗
Figure 3
Figure 3. Data Selection on PASCAL VOC. For object detection we use a FasterRCNN with ResNet-18 backbone and for semantic segmentation a Segmenter with ViT-T backbone where the decoder part of the models is trained from scratch. OFDS consistently performs among the best methods on PASCAL VOC and outperforms them on PASCAL VOC with class imbalance (class imbal.). 0.1 0.2 0.3 25 30 35 mAP Full Dataset Human Labels: 40.1 Autolab… view at source ↗
Figures from the paper (26 more)
Figure 4
Figure 4. Figure 4: Data Selection on Cityscapes. The results are obtained with the same models as for [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 4
Figure 4. Figure 4: Cityscapes naturally has a more imbalanced class distribution than PASCAL VOC (see App. H). We observe that OFDS reliably performs among the best methods. The performance of OFDS is especially notable when evaluating on the rarest classes which highlights its ability t…
Figure 5
Figure 5. Figure 5: Data Selection on LoveDA. The setup is the same as the segmentation results from [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Combining Autolabels With Data Selec￾tion Improves Downstream Performance under Con￾strained Annotation Budgets. Dashed: models trained as in Sec. 4.2. Solid: fine-tuning the checkpoint pre￾trained with autolabels on selected subsets with human labels. Results for City…
Figure 7
Figure 7. Figure 7: Initial Data Selection Through OFDS Improves Active Learning. We train a ResNet-50 RetinaNet [29]for detection using PPAL [57] and Wide-ResNet38 DeepLabv3+ [6] for semantic segmentation using EquAL. The only differ￾ence is the selection of the initial data set. As disc…
Figure 8
Figure 8. Figure 8: Ablations with Different Models for the Downstream Tasks. OFDS also leads to the best results for training a Deformable DETR for object detection and a PSP Net for semantic segmentation. 1 [PITH_FULL_IMAGE:figures/full_fig_p014_8.png]
Figure 9
Figure 9. Figure 9: OFDS Outperforms Image-Focused Baseline. The image-focused baseline is based on CLIP retrieval with an evenly split budget between all classes. The model and training hyperparame￾ters are the same as in [PITH_FULL_IMAGE:figures/full_fig_p015_9.png]
Figure 10
Figure 10. Figure 10: Images with High Text-to-Image Similarity for Class ”Train” in Cityscapes. We use a SigLip 2 ViT-B/16 CLIP model to perform retrieval on Cityscapes. The depicted images are among the top 25 images with the highest text-to-image similarity for the class ”train” of whic…
Figure 11
Figure 11. Figure 11 [PITH_FULL_IMAGE:figures/full_fig_p016_11.png]
Figure 12
Figure 12. Figure 12: Illustration of Clusters for Classes from PASCAL VOC. The clustering was performed with NCl = 30 clusters per class. The resulting clusters exhibit the semantic consistency which we use to construct a uniform covering of the target classes. C.2 Cluster Illustrations I…
Figure 13
Figure 13. Figure 13: Leveraging Autolabels Together with OFDS further improves the Cold Start Problem. The models are trained with the same setup as for [PITH_FULL_IMAGE:figures/full_fig_p018_13.png]
Figure 14
Figure 14. Figure 14: Different Initial Labeled Datasets for Active Semantic Segmentation on PASCAL VOC with Class Imbalance. We train a DeepLavV3+ model with Wide-ResNet38 backbone on PASCAL VOC with class imbalance using the setup from Mittal et al. [34]. For every plot, the models were …
Figure 15
Figure 15. Figure 15: Different Initial Labeled Datasets for Active Semantic Segmentation on Cityscapes. We use the same setup as in [PITH_FULL_IMAGE:figures/full_fig_p019_15.png]
Figure 16
Figure 16. Figure 16: Direct Comparison of OFDS and Active Learning. We compare the models trained with the active learning frameworks from [PITH_FULL_IMAGE:figures/full_fig_p020_16.png]
Figure 17
Figure 17. Figure 17: OFDS Selects More Class-Balanced Subsets. We compute the class balance score intro￾duced by Sorscher et al. [45] and find that the subsets selected by OFDS consistently feature higher class balance scores than the baselines. This indicates more evenly distributed clas…
Figure 18
Figure 18. Figure 18: Repeated Data Selection For Baselines With Randomness. Since the random baseline and FreeSel are based on probabilistic selection, we repeat the data selection process for object detection and semantic segmentation on PASCAL VOC with class imbalance three times. The s…
Figure 19
Figure 19. Figure 19: Using a Stronger Object Proposer Further Improves OFDS. We exchange the Ground￾ing SAM2 tiny model with the Grounding SAM2 base model which features a larger architecture and was trained on more data. In Section 4, we employ Grounding SAM2 tiny to provide a fair compa…
Figure 20
Figure 20. Figure 20: Class distributions for the Object Detection Datasets. trainbus truck motorcyclewall rider fence sky road bicycle terrain sidewalk building traffic light vegetation person traffic sign car pole Classes 0 10000 20000 30000 40000 Number of Units Class Distribution for C…
Figure 21
Figure 21. Figure 21: Class distributions for the Semantic Segmentation Datasets. agricultural field building water surface road barren area forest Classes 0 1000 2000 Number of Units [PITH_FULL_IMAGE:figures/full_fig_p023_21.png]
Figure 22
Figure 22. Figure 22: Class distributions for the LoveDA dataset. 10 [PITH_FULL_IMAGE:figures/full_fig_p023_22.png]
Figure 23
Figure 23. Figure 23: Calibration of the Object Proposer using the FPR to control the precision for OFDS and the F1 score for autolabeling. The calibration data is as described in Section I.1 evaluate the performance on the five smallest classes, accounting for one fourth of the total clas…
Figure 24
Figure 24. Figure 24: Compute Time for the Data Selection in OFDS Without the Time to Generate the Object Proposals The clustering and selection was performed on a Xeon Gold 6150 CPU. 0.05 0.10 0.15 0.20 0.25 0.30 Fraction of Dataset by Number of Units 22.5 25.0 27.5 30.0 32.5 35.0 mAP Ful…
Figure 25
Figure 25. Figure 25: Combining Autolabels With Data Selection on Cityscapes. The training setup is the same as for [PITH_FULL_IMAGE:figures/full_fig_p026_25.png]
Figure 26
Figure 26. Figure 26: Complete Results for Combining Autolabels with Data Selection on PASCAL VOC with Class Imbalance. The results correspond to a FasterRCNN with ResNet-18 backbone and a Segmenter with ViT-T backbone. The models were first pre-trained on the full dataset with autolabels …
Figure 27
Figure 27. Figure 27: Complete Results for Combining Autolabels with Data Selection on Cityscapes. The results were obtained using the same setup as for Figures 6 and 26. 0.05 0.10 0.15 0.20 0.25 0.30 Fraction of Dataset by Number of Units 0.12 0.14 0.16 0.18 0.20 0.22 0.24 mAP Instance Se…
Figure 28
Figure 28. Figure 28: Results for Instance Segmentation on Cityscapes. The downstream models trained for the tasks is a MaskRCNN with ResNet-18 backbone. OFDS consistently outperforms all baselines also for this task. 15 [PITH_FULL_IMAGE:figures/full_fig_p028_28.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

67 extracted references · 47 canonical work pages

  1. [1]

    Amro Abbas, Evgenia Rusak, Kushal Tirumala, Wieland Brendel, Kamalika Chaudhuri, and Ari S. Morcos. Effective pruning of web-scale datasets based on complexity of concept clusters. In ICLR, 2024

    work page 2024

  2. [2]

    Semd- edup: Data-efficient learning at web-scale through semantic deduplication

    Amro Abbas, Kushal Tirumala, Dániel Simig, Surya Ganguli, and Ari S Morcos. Semd- edup: Data-efficient learning at web-scale through semantic deduplication. arXiv preprint arXiv:2303.09540, 2023

    arXiv 2023

  3. [3]

    Grua, Pepijn van de Ven, and Ciarán Eising

    Cathaoir Agnew, Anthony Scanlan, Patrick Denny, Eoin M. Grua, Pepijn van de Ven, and Ciarán Eising. Annotation quality versus quantity for object detection and instance segmentation. IEEE Access, 12:140958–140977, 2024

    work page 2024

  4. [4]

    Emerging properties in self-supervised vision transformers

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

    2021

  5. [5]

    MMDetection: Open mmlab detection toolbox and benchmark

    Kai Chen, Jiaqi Wang, Jiangmiao Pang, Yuhang Cao, Yu Xiong, Xiaoxiao Li, Shuyang Sun, Wansen Feng, Ziwei Liu, Jiarui Xu, Zheng Zhang, Dazhi Cheng, Chenchen Zhu, Tianheng Cheng, Qijie Zhao, Buyu Li, Xin Lu, Rui Zhu, Yue Wu, Jifeng Dai, Jingdong Wang, Jianping Shi, Wanli Ouyang, Chen Change Loy, and Dahua Lin. MMDetection: Open mmlab detection toolbox and b...

    arXiv 1906

  6. [6]

    Encoder-decoder with atrous separable convolution for semantic image segmentation

    Liang-Chieh Chen, Yukun Zhu, George Papandreou, Florian Schroff, and Hartwig Adam. Encoder-decoder with atrous separable convolution for semantic image segmentation. In ECCV, 2018

    2018

  7. [7]

    Making your first choice: To address cold start problem in medical active learning

    Liangyu Chen, Yutong Bai, Siyu Huang, Yongyi Lu, Bihan Wen, Alan Yuille, and Zongwei Zhou. Making your first choice: To address cold start problem in medical active learning. In MIDL, 2024

    work page 2024

  8. [8]

    Mmsegmentation: Openmmlab semantic segmentation toolbox and benchmark

    MMSegmentation Contributors. Mmsegmentation: Openmmlab semantic segmentation toolbox and benchmark. https://github.com/open-mmlab/mmsegmentation, 2020

    work page 2020

Show all 67 references
  1. [9]

    The cityscapes dataset for semantic urban scene understanding

    Marius Cordts, Mohamed Omran, Sebastian Ramos, Timo Rehfeld, Markus Enzweiler, Rodrigo Benenson, Uwe Franke, Stefan Roth, and Bernt Schiele. The cityscapes dataset for semantic urban scene understanding. In CVPR, 2016

    2016

  2. [10]

    Learning from partial labels

    Timothee Cour, Ben Sapp, and Ben Taskar. Learning from partial labels. JMLR, 12(42):1501– 1536, 2011

    2011

  3. [11]

    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

    2009

  4. [12]

    Boosting long-tailed object detection via step-wise learning on smooth-tail data

    Na Dong, Yongqiang Zhang, Mingli Ding, and Gim Hee Lee. Boosting long-tailed object detection via step-wise learning on smooth-tail data. In ICCV, 2023

    2023

  5. [13]

    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, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition at...

    2021

  6. [14]

    Active learning at the imagenet scale

    Zeyad Ali Sami Emam, Hong-Min Chu, Ping-Yeh Chiang, Wojciech Czaja, Richard Leapman, Micah Goldblum, and Tom Goldstein. Active learning at the imagenet scale. arXiv preprint arXiv:2111.12880, 2021

    2021 arXiv

  7. [15]

    Williams, John Winn, and Andrew Zisserman

    Mark Everingham, Luc Gool, Christopher K. Williams, John Winn, and Andrew Zisserman. The pascal visual object classes (voc) challenge. IJCV, 88(2):303–338, 2010

    2010

  8. [16]

    Introduction to core-sets: an updated survey

    Dan Feldman. Introduction to core-sets: an updated survey. arXiv preprint arXiv:2011.09384, 2020

    2011 arXiv

  9. [17]

    Alireza Golestaneh and Kris Kitani

    S. Alireza Golestaneh and Kris Kitani. Importance of self-consistency in active learning for semantic segmentation. BMVC, 2020

    2020

  10. [18]

    Deepcore: A comprehensive library for coreset selection in deep learning

    Chengcheng Guo, Bo Zhao, and Yanbing Bai. Deepcore: A comprehensive library for coreset selection in deep learning. In Database and Expert Systems Applications , 2022

    2022

  11. [19]

    Zhang, Shaoqing Ren, and Jian Sun

    Kaiming He, X. Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In CVPR, 2011

    2011

  12. [20]

    Alex Holub, Pietro Perona, and Michael C. Burl. Entropy-based active learning for object recognition. In CVPR workshops, 2008

    2008

  13. [21]

    Active learning for semantic segmentation with multi-class label query

    Sehyun Hwang, Sohyun Lee, Hoyoung Kim, Minhyeon Oh, Jungseul Ok, and Suha Kwak. Active learning for semantic segmentation with multi-class label query. In NeurIPS, 2023

    2023

  14. [22]

    Completr: Reducing the cost of annotations for object detection in dense scenes with vision transformers

    Achin Jain, Kibok Lee, Gurumurthy Swaminathan, Hao Yang, Bernt Schiele, Avinash Ravichan- dran, and Onkar Dabeer. Completr: Reducing the cost of annotations for object detection in dense scenes with vision transformers. arXiv preprint arXiv:2209.05654, 2022

    2022 arXiv

  15. [23]

    Johnson and Taghi M

    Justin M. Johnson and Taghi M. Khoshgoftaar. Survey on deep learning with class imbalance. Journal of Big Data, 6:1–54, 2019

    2019

  16. [24]

    Active label correction for semantic segmentation with foundation models

    Hoyoung Kim, Sehyun Hwang, Suha Kwak, and Jungseul Ok. Active label correction for semantic segmentation with foundation models. In ICML, 2024

    2024

  17. [25]

    Coreset selection for object detection

    Hojun Lee, Suyoung Kim, Junhoo Lee, Jaeyoung Yoo, and Nojun Kwak. Coreset selection for object detection. In CVPR Workshops, 2024

    2024

  18. [26]

    Blip-2: bootstrapping language-image pre-training with frozen image encoders and large language models

    Junnan Li, Dongxu Li, Silvio Savarese, and Steven Hoi. Blip-2: bootstrapping language-image pre-training with frozen image encoders and large language models. In ICML, 2023

    2023

  19. [27]

    UP-DP: Unsupervised prompt learning for data pre-selection with vision-language models

    Xin Li, Sima Behpour, Thang Doan, Wenbin He, Liang Gou, and Liu Ren. UP-DP: Unsupervised prompt learning for data pre-selection with vision-language models. In NeurIPS, 2023

    2023

  20. [28]

    Block annotation: Better image annotation with sub-image decomposition

    Hubert Lin, Paul Upchurch, and Kavita Bala. Block annotation: Better image annotation with sub-image decomposition. In ICCV, 2019

    2019

  21. [29]

    Focal loss for dense object detection

    Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollár. Focal loss for dense object detection. In ICCV, 2017

    2017

  22. [30]

    Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C

    Tsung-Yi Lin, Michael Maire, Serge J. Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C. Lawrence Zitnick. Microsoft coco: Common objects in context. In ECCV, 2014

    2014

  23. [31]

    Grounding dino: Marrying dino with grounded pre-training for open-set object detection

    Shilong Liu, Zhaoyang Zeng, Tianhe Ren, Feng Li, Hao Zhang, Jie Yang, Chunyuan Li, Jianwei Yang, Hang Su, Jun Zhu, et al. Grounding dino: Marrying dino with grounded pre-training for open-set object detection. In ECCV, 2024

    2024

  24. [32]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. In ICLR, 2019

    2019

  25. [33]

    Coresets for data-efficient training of machine learning models

    Baharan Mirzasoleiman, Jeff Bilmes, and Jure Leskovec. Coresets for data-efficient training of machine learning models. In ICML, 2020

    2020

  26. [34]

    Schäfer, and Thomas Brox

    Sudhanshu Mittal, Joshua Niemeijer, Jörg P. Schäfer, and Thomas Brox. Best practices in active learning for semantic segmentation. In GCPR, 2023. 11

    2023

  27. [35]

    Roth, and Daguang Xu

    Vishwesh Nath, Dong Yang, Holger R. Roth, and Daguang Xu. Warm start active learning with proxy labels and selection via semi-supervised fine-tuning. In MICCAI, 2022

    2022

  28. [36]

    Activeglae: A benchmark for deep active learning with transformers

    Lukas Rauch, Matthias Aßenmacher, Denis Huseljic, Moritz Wirth, Bernd Bischl, and Bernhard Sick. Activeglae: A benchmark for deep active learning with transformers. In ECML PKDD 2023, 2023

    2023

  29. [37]

    Sam 2: Segment anything in images and videos

    Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Roman Rädle, Chloe Rolland, Laura Gustafson, Eric Mintun, Junting Pan, Kalyan Vasudev Alwala, Nicolas Carion, Chao-Yuan Wu, Ross Girshick, Piotr Dollár, and Christoph Feichtenh...

    2024 arXiv

  30. [38]

    Gupta, Xiaojiang Chen, and Xin Wang

    Pengzhen Ren, Yun Xiao, Xiaojun Chang, Po-Yao Huang, Zhihui Li, Brij B. Gupta, Xiaojiang Chen, and Xin Wang. A survey of deep active learning. ACM Comput. Surv., 54(9), 2021

    2021

  31. [39]

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

    Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real-time object detection with region proposal networks. In C. Cortes, N. Lawrence, D. Lee, M. Sugiyama, and R. Garnett, editors, NeurIPS, 2015

    2015

  32. [40]

    Grounded sam: Assembling open-world models for diverse visual tasks

    Tianhe Ren, Shilong Liu, Ailing Zeng, Jing Lin, Kunchang Li, He Cao, Jiayu Chen, Xinyu Huang, Yukang Chen, Feng Yan, Zhaoyang Zeng, Hao Zhang, Feng Li, Jie Yang, Hongyang Li, Qing Jiang, and Lei Zhang. Grounded sam: Assembling open-world models for diverse visual tasks. arXiv ...

    2024 arXiv

  33. [41]

    You never get a second chance to make a good first impression: Seeding active learning for 3d semantic segmentation

    Nermin Samet, Oriane Siméoni, Gilles Puy, Georgy Ponimatkin, Renaud Marlet, and Vincent Lepetit. You never get a second chance to make a good first impression: Seeding active learning for 3d semantic segmentation. In ICCV, 2023

    2023

  34. [42]

    Active learning for convolutional neural networks: A core-set approach

    Ozan Sener and Silvio Savarese. Active learning for convolutional neural networks: A core-set approach. In ICLR, 2018

    2018

  35. [43]

    C. E. Shannon. A mathematical theory of communication. The Bell System Technical Journal, 1948

    1948

  36. [44]

    Reco: Retrieve and co-segment for zero-shot transfer

    Gyungin Shin, Weidi Xie, and Samuel Albanie. Reco: Retrieve and co-segment for zero-shot transfer. In NeurIPS, 2022

    2022

  37. [45]

    Ben Sorscher, Robert Geirhos, Shashank Shekhar, Surya Ganguli, and Ari S. Morcos. Beyond neural scaling laws: beating power law scaling via data pruning. In NeurIPS, 2022

    2022

  38. [46]

    Segmenter: Transformer for semantic segmentation

    Robin Strudel, Ricardo Garcia, Ivan Laptev, and Cordelia Schmid. Segmenter: Transformer for semantic segmentation. In ICCV, 2021

    2021

  39. [47]

    Equalization loss v2: A new gradient balance approach for long-tailed object detection

    Jingru Tan, Xin Lu, Gang Zhang, Changqing Yin, and Quanquan Li. Equalization loss v2: A new gradient balance approach for long-tailed object detection. In CVPR, 2021

    2021

  40. [48]

    Siglip 2: Multilingual vision- language encoders with improved semantic understanding, localization, and dense features

    Michael Tschannen, Alexey Gritsenko, Xiao Wang, Muhammad Ferjad Naeem, Ibrahim Alab- dulmohsin, Nikhil Parthasarathy, Talfan Evans, Lucas Beyer, Ye Xia, Basil Mustafa, Olivier Hénaff, Jeremiah Harmsen, Andreas Steiner, and Xiaohua Zhai. Siglip 2: Multilingual vision- language ...

    2025 arXiv

  41. [49]

    Seesaw loss for long-tailed instance segmentation

    Jiaqi Wang, Wenwei Zhang, Yuhang Zang, Yuhang Cao, Jiangmiao Pang, Tao Gong, Kai Chen, Ziwei Liu, Chen Change Loy, and Dahua Lin. Seesaw loss for long-tailed instance segmentation. In CVPR, 2021

    2021

  42. [50]

    Loveda: A remote sensing land-cover dataset for domain adaptive semantic segmentation

    Junjue Wang, Zhuo Zheng, Ailong Ma, Xiaoyan Lu, and Yanfei Zhong. Loveda: A remote sensing land-cover dataset for domain adaptive semantic segmentation. In NeurIPS, 2021

    2021

  43. [51]

    The devil is in classification: A simple framework for long-tail instance segmentation

    Tao Wang, Yu Li, Bingyi Kang, Junnan Li, Junhao Liew, Sheng Tang, Steven Hoi, and Jiashi Feng. The devil is in classification: A simple framework for long-tail instance segmentation. In ECCV, 2020. 12

    2020

  44. [52]

    Unsupervised selective labeling for more effective semi-supervised learning

    Xudong Wang, Long Lian, and Stella X Yu. Unsupervised selective labeling for more effective semi-supervised learning. In ECCV, 2022

    2022

  45. [53]

    On the de-duplication of laion-2b

    Ryan Webster, Julien Rabin, Loic Simon, and Frederic Jurie. On the de-duplication of laion-2b. arXiv preprint arXiv:2303.12733, 2023

    2023 arXiv

  46. [54]

    Jingqian Wu, Rongtao Xu, Zach Wood-Doughty, Changwei Wang, Shibiao Xu, and Edmund Y . Lam. Segment anything model is a good teacher for local feature learning. arXiv preprint arXiv:2309.16992, 2024

    2024

  47. [55]

    Towards free data selection with general-purpose models

    Yichen Xie, Mingyu Ding, Masayoshi Tomizuka, and Wei Zhan. Towards free data selection with general-purpose models. In NeurIPS, 2023

    2023

  48. [56]

    Missing labels in object detection

    Mengmeng Xu, Yancheng Bai, and Bernard Ghanem. Missing labels in object detection. In CVPR Workshops, 2019

    2019

  49. [57]

    Chenhongyi Yang, Lichao Huang, and Elliot J. Crowley. Plug and Play Active Learning for Object Detection. In CVPR, 2024

    2024

  50. [58]

    Large-scale training data search for object re- identification

    Yue Yao, Tom Gedeon, and Liang Zheng. Large-scale training data search for object re- identification. In CVPR, 2023

    2023

  51. [59]

    Wide residual networks

    Sergey Zagoruyko and Nikos Komodakis. Wide residual networks. In BMVC, 2016

    2016

  52. [60]

    Data-centric artificial intelligence: A survey

    Daochen Zha, Zaid Pervaiz Bhat, Kwei-Herng Lai, Fan Yang, Zhimeng Jiang, Shaochen Zhong, and Xia Hu. Data-centric artificial intelligence: A survey. ACM Comput. Surv., 57(5), 2025

    2025

  53. [61]

    Alps: An auto-labeling and pre-training scheme for remote sensing segmentation with segment anything model

    Song Zhang, Qingzhong Wang, Junyi Liu, and Haoyi Xiong. Alps: An auto-labeling and pre-training scheme for remote sensing segmentation with segment anything model. arXiv preprint arXiv:2406.10855, 2024

    2024 arXiv

  54. [62]

    Mazurowski

    Yixin Zhang, Shen Zhao, Hanxue Gu, and Maciej A. Mazurowski. How to efficiently annotate images for best-performing deep learning based segmentation models: An empirical study with weak and noisy annotations and segment anything model. arXiv preprint arXiv:2312.10600, 2023

    2023 arXiv

  55. [63]

    Pyramid scene parsing network

    Hengshuang Zhao, Jianping Shi, Xiaojuan Qi, Xiaogang Wang, and Jiaya Jia. Pyramid scene parsing network. In CVPR, 2017

    2017

  56. [64]

    Bartoldson, Bhavya Kailkhura, and Atul Prakash

    Haizhong Zheng, Elisa Tsai, Yifu Lu, Jiachen Sun, Brian R. Bartoldson, Bhavya Kailkhura, and Atul Prakash. Elfs: Label-free coreset selection with proxy training dynamics. In ICLR, 2025

    2025

  57. [65]

    Optimizing object detection via metric-driven training data selection

    Changyuan Zhou, Yumin Guo, Qinxue Lv, and Ji Yuan. Optimizing object detection via metric-driven training data selection. In CVPR Workshops, 2024

    2024

  58. [66]

    Deformable detr: Deformable transformers for end-to-end object detection

    Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, and Jifeng Dai. Deformable detr: Deformable transformers for end-to-end object detection. In ICLR, 2021. 13 Appendix We start with an overview of the content of the Appendix: • In Section A, we perform ablations with diff...

    2021

  59. [67]

    a photo of a {classname}

    and the random baseline. Entropy and coreset are two of the most commonly used frameworks for active learning and remain strong baselines even in more recent works [57]. Coreset for active learning differs from K-Centers for data selection from Section 3.2. Coreset for active ...

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.