Pith. sign in

REVIEW 2 major objections 4 minor 57 references

A hybrid of text-to-image generation and verified image-to-image editing lifts rare-class instance segmentation on LVIS by up to 9.5 AP points.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-10 11:24 UTC pith:62KFVVMJ

load-bearing objection Solid hybrid synthesis recipe that actually moves rare-class LVIS numbers; the VRAIN residual-error worry is real but secondary to the ablations. the 2 major comments →

arxiv 2607.08201 v1 pith:62KFVVMJ submitted 2026-07-09 cs.CV cs.AI

TMI: Text-to-Image Meets Image-to-Image for Complementary Data Synthesis to Boost Long-Tailed Instance Segmentation

Hyeonseop Song , Seokhun Choi , Hoseok Do This is my paper
classification cs.CV cs.AI
keywords long-tailed instance segmentationtext-to-image synthesisimage-to-image editingVRAINteacher-student pseudo-labelingLVISgenerative data augmentation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Large-vocabulary instance segmentation is held back by long-tailed category counts: rare objects appear so seldom that models never learn them well, and synthetic data has so far been only a partial fix. Pure text-to-image pipelines create diverse scenes but produce noisy labels, while copy-paste methods give clean labels yet break scene realism. This paper shows that the two can be made complementary. Text-to-image generation supplies broad scene and category coverage, with a teacher-student loop that keeps only prompt-consistent labels and adapts them online. A new image-to-image editor called VRAIN then inserts rare-class objects into real photographs at locations a vision-language model judges natural, and verifies each insertion before accepting its mask. On the LVIS benchmark the combined synthetic set raises overall box AP by as much as 4 points and rare-class AP by as much as 9.5 points over strong real-only and synthetic baselines, and the gains grow with larger backbones. The result is a practical recipe for turning modern generative models into reliable training data for the long tail.

Core claim

Coupling text-to-image generation that is cleaned by prompt-consistent filtering and teacher-student adaptation with a verified, instruction-based image-to-image editor that places rare objects into real scenes yields synthetic training data that simultaneously improves overall and rare-class instance segmentation on LVIS, outperforming either paradigm alone.

What carries the argument

VRAIN, a place-and-verify I2I pipeline: a vision-language model proposes a rare class and a natural-language placement instruction; an instruction editor synthesizes the edit; SSIM, open-vocabulary detection, a second VLM check, and SAM then confirm semantic fidelity and produce a trustworthy mask that is merged with the original annotation.

Load-bearing premise

The verification loop must keep residual false positives and domain gaps small enough that the rare-class images can bootstrap reliable teacher labels on pure text-to-image data; if it systematically accepts subtle mismatches or rejects valid rare instances, the rare-class gains disappear.

What would settle it

Generate the same volume of I2I data with the VLM verification stage turned off (or replaced by a weaker filter) and retrain under identical budgets; if rare-class AP no longer rises above the real-only or copy-paste baselines, the claim that verified placement is essential is falsified.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 4 minor

Summary. The paper proposes TMI, a hybrid synthetic-data pipeline for long-tailed instance segmentation that couples text-to-image (T2I) generation with a context-aware image-to-image editor called VRAIN. T2I images are produced from GPT-4o prompts over random LVIS category subsets and filtered by text-consistent offline labels; an EMA teacher-student loop then adaptively refines those labels online, treating high-confidence prompt-inconsistent detections as unlabeled localization targets. VRAIN inserts rare-class instances into real LVIS images via VLM-proposed instructions, Flux-Kontext editing, SSIM localization, open-vocabulary detection, red-box VLM verification, and SAM masking, yielding high-fidelity DI2I that bootstraps rare-class supervision. On LVIS validation with CenterNet2, the hybrid data raise APbox from 34.5 to 38.1 (ResNet-50) and 47.5 to 50.7 (Swin-L), with rare-class gains of +9.9 and +7.7 respectively, outperforming MosaicFusion, X-Paste and DiverGen under matched budgets.

Significance. If the residual error of the VRAIN place-and-verify loop is as low as claimed, the work supplies a practical, scalable recipe that simultaneously addresses the complementary failure modes of pure T2I (label noise, rare-class weakness) and pure copy-paste (contextual artifacts). The systematic ablations (Tables 1, 3–7 and supplementary Tables A–C) isolate filtering, progressive versus frozen teacher, unlabeled localization, VLM verification and data scale, and the rare-targeted DiverGen negative control is informative. The approach scales with backbone capacity and is built on publicly available generative models, making the gains potentially useful for large-vocabulary instance segmentation beyond LVIS.

major comments (2)
  1. The central rare-class claim (Table 2: +9.9 / +7.7 APbox_r) rests on the assumption that VRAIN’s residual false-positive rate under LVIS fine-grained ambiguity is low enough for the EMA teacher to bootstrap reliable rare labels on pure T2I images (Sec. 3.2, Fig. 3). Tables 5–6 and the ORIDa 0.925 mAP check only measure coarse class-label correctness and mask fidelity on controlled/perturbed data; they do not quantify residual confusions among LVIS rare classes (e.g., headset vs helmet, phonebook vs phone) that survive the binary VLM filter. The paper’s own limitation section acknowledges imperfect instruction adherence, yet no end-to-end error rate of the full pipeline on LVIS rare categories is reported. Without that measurement (or a human audit of a DI2I subset), it remains possible that the rare-class gains partly reflect residual label noise rather than true representation learning—
  2. Several free thresholds (τ_label = 0.7, τ_unlabel = 1.2, τ_edit = 5, EMA γ = 0.999, Q = 5, N_T2I / N_I2I) are fixed without sensitivity analysis (Sec. 4.1). Because the online adaptive threshold and the unlabeled-localization rule are load-bearing for the teacher-student loop (Sec. 3.3, supplementary Tables B–C), at least a modest sweep or stability plot is needed to establish that the reported gains are not brittle to these choices.
minor comments (4)
  1. Fig. 1 caption and the abstract claim “up to +9.5” rare AP while Table 2 reports +9.9 (ResNet-50 APbox_r); reconcile the numbers.
  2. The offline labeler MP is cited only as [57] (Co-DETR); state the exact checkpoint and training data so that the text-consistent filtering baseline is reproducible.
  3. Supplementary Fig. B reports a 52 % acceptance rate for VRAIN; move a short summary of rejection modes into the main text (Sec. 3.2 or 4.3) so readers can judge quality–quantity trade-offs without the supplement.
  4. Notation for the merged supervision Am (Sec. 3.3) is dense; a short algorithmic box or pseudocode would clarify the IoU-matching and adaptive-threshold steps.

Circularity Check

0 steps flagged

No circularity: LVIS AP gains are measured on an external held-out validation set against independently published baselines; synthetic labels and free thresholds do not define the reported metrics by construction.

full rationale

The paper's central claims are empirical performance numbers (Table 2: APbox 34.5 o38.1 ResNet-50, 47.5 o50.7 Swin-L; rare-class gains up to +9.9) obtained by training CenterNet2 on a mixture of real LVIS training images plus synthetically generated DT2I and DI2I, then evaluating on the official LVIS validation set. The evaluation metric is standard COCO-style AP on real images and is independent of the synthetic labels, the EMA teacher, the VLM verification filter, or the hand-chosen thresholds (τ_label=0.7, τ_unlabel=1.2, τ_edit=5). Ablations (Tables 1, 3–7) and the ORIDa mask-fidelity check further isolate components without redefining the target. There are no self-definitional equations, no fitted parameters renamed as predictions of the same quantity, no load-bearing uniqueness theorems imported from the authors' prior work, and no renaming of known empirical patterns. The derivation chain is therefore an ordinary engineering pipeline whose success or failure is externally falsifiable; circularity score is zero.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 1 invented entities

The central claim rests on standard CV assumptions (LVIS frequency splits, CenterNet2 training protocol, diffusion models as black-box generators) plus several hand-chosen thresholds and the unproved premise that VLM verification is sufficiently accurate for rare LVIS classes. No new physical entities are postulated; VRAIN is an algorithmic pipeline.

free parameters (5)
  • tau_label = 0.7
    Class-wise adaptive threshold multiplier for prompt-consistent teacher predictions; set to 0.7 without sweep reported in main text.
  • tau_unlabel = 1.2
    Multiplier for treating high-confidence prompt-inconsistent detections as unlabeled; set to 1.2.
  • tau_edit = 5
    SSIM difference threshold that defines the edited region for open-vocab detection; set to 5.
  • EMA decay gamma = 0.999
    Teacher update rate; fixed at 0.999 following prior semi-supervised practice.
  • N_T2I / N_I2I / Q / l = 200k / 80k / 5 / [5,10]
    Data volumes (200k T2I, 80k I2I), candidate set size Q=5, and categories per prompt l in [5,10] are design choices that affect final AP.
axioms (4)
  • domain assumption Offline public instance segmentor MP produces usable initial pseudo-labels on Flux-generated images after text-consistent filtering.
    Sec. 3.1; Table 1 shows the filter helps but residual noise remains.
  • domain assumption InternVL3-14B can both propose contextually appropriate rare-class insertions and later verify them with low false-positive rate.
    Sec. 3.2 and Table 5; acceptance rate ~52% implies heavy filtering.
  • standard math LVIS rare/common/frequent splits and CenterNet2 training schedule are the correct evaluation protocol for fair comparison.
    Standard in the cited baselines; Sec. 4.1.
  • ad hoc to paper Prompt-inconsistent high-confidence detections should contribute only localization (box/mask) loss, not classification loss.
    Sec. 3.3 and Table C; design choice validated by ablation but not forced by theory.
invented entities (1)
  • VRAIN place-and-verify pipeline independent evidence
    purpose: Generate high-fidelity rare-class annotations by instruction editing followed by multi-stage verification.
    Algorithmic construct, not a physical entity; independent evidence is the ORIDa mask mAP 0.925 and the LVIS rare-class gains when verification is ablated.

pith-pipeline@v1.1.0-grok45 · 26615 in / 3122 out tokens · 33150 ms · 2026-07-10T11:24:40.124938+00:00 · methodology

0 comments
read the original abstract

Large-vocabulary instance segmentation is constrained by long-tailed category distributions and fine-grained inter-class ambiguity. While data synthesis offers a promising alternative, current paradigms have complementary limitations: text-to-image (T2I) methods inherit noisy pseudo-labels and struggle on rare classes, whereas copy-paste methods compromise contextual realism. To address these issues, we propose a hybrid pipeline coupling T2I generation with context-aware image-to-image (I2I) editing. The T2I branch provides broad category and scene diversity, while a teacher-student scheme ensures label reliability by selectively retaining only prompt-specified categories. To strengthen supervision for rare classes, we introduce VRAIN (Verified Rare-class Augmentation via INstructed editing), a novel I2I editor. VRAIN inserts high-confidence instances at semantically appropriate locations within in-the-wild scenes, yielding semantically coherent and visually natural edits that reduce domain gaps and enable targeted augmentation. On the LVIS benchmark, our method surpasses existing baselines, improving overall AP by up to +4.0 points and rare-class AP by up to +9.5 points, while scaling effectively with backbone capacity. Our project page is available at https://seokhunchoi.github.io/TMI

Figures

Figures reproduced from arXiv: 2607.08201 by Hoseok Do, Hyeonseop Song, Seokhun Choi.

Figure 1
Figure 1. Figure 1: Mutually Complementary T2I-I2I Data Synthesis. (a) [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of Hybrid Data Generation and Training Framework. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: VRAIN “Place-and-Verify” Pipeline. Our two-stage framework ensures high-fidelity I2I editing. (i) Place: A VLM proposes a semantically coherent instruc￾tion (inst∗ ) for inserting a rare class (c ∗ ) that fits naturally within Ireal. An instruction￾based editor ΦI2I then synthesizes Iedit. (ii) Verify: Iedit is then validated. The new instance is localized via SSIM difference and open-vocab detection, sema… view at source ↗
Figure 4
Figure 4. Figure 4: Examples of Pseudo-label Adaptation in T2I images. [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Examples of Final Pseudo-labels for T2I images. [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative Examples of the VRAIN Verification Stage. [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

57 extracted references · 57 canonical work pages · 11 internal anchors

  1. [1]

    YOLOv4: Optimal Speed and Accuracy of Object Detection

    Bochkovskiy, A., Wang, C.Y., Liao, H.Y.M.: Yolov4: Optimal speed and accuracy of object detection. arXiv preprint arXiv:2004.10934 (2020)

    work page internal anchor Pith review Pith/arXiv arXiv 2004

  2. [2]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Brooks, T., Holynski, A., Efros, A.A.: Instructpix2pix: Learning to follow image editing instructions. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 18392–18402 (2023)

    work page 2023

  3. [3]

    In: International conference on machine learning

    Chang, N., Yu, Z., Wang, Y.X., Anandkumar, A., Fidler, S., Alvarez, J.M.: Image- level or object-level? a tale of two resampling strategies for long-tailed detection. In: International conference on machine learning. pp. 1463–1472. PMLR (2021)

    work page 2021

  4. [4]

    In: Proc

    Cordts, M., Omran, M., Ramos, S., Rehfeld, T., Enzweiler, M., Benenson, R., Franke, U., Roth, S., Schiele, B.: The cityscapes dataset for semantic urban scene understanding. In: Proc. of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2016)

    work page 2016

  5. [5]

    In: 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

    De Brabandere, B., Neven, D., Van Gool, L.: Semantic instance segmentation for autonomous driving. In: 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). pp. 478–480 (2017).https://doi.org/10. 1109/CVPRW.2017.66

    work page 2017

  6. [6]

    In: Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition

    Fan, C., Zhu, M., Chen, H., Liu, Y., Wu, W., Zhang, H., Shen, C.: Divergen: Improving instance segmentation by learning wider data distribution with more diverse generative data. In: Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition. pp. 3986–3995 (2024)

    work page 2024

  7. [7]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Fu, S., Yang, Q., Mo, Q., Yan, J., Wei, X., Meng, J., Xie, X., Zheng, W.S.: Llmdet: Learning strong open-vocabulary object detectors under the supervision of large language models. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 14987–14997 (2025)

    work page 2025

  8. [8]

    In: Computer Vision – ECCV 2024: 18th European Conference, Milan, Italy, September 29–October 4, 2024, Proceedings, Part XLVIII

    Guo, H., Zhu, H., Peng, S., Wang, Y., Shen, Y., Hu, R., Zhou, X.: Sam-guided graph cut for 3d instance segmentation. In: Computer Vision – ECCV 2024: 18th European Conference, Milan, Italy, September 29–October 4, 2024, Proceedings, Part XLVIII. p. 234–251. Springer-Verlag, Berlin, Heidelberg (2024).https:// doi.org/10.1007/978-3-031-73195-2_14,https://do...

    work page doi:10.1007/978-3-031-73195-2_14 2024

  9. [9]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Gupta, A., Dollar, P., Girshick, R.: Lvis: A dataset for large vocabulary instance segmentation. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 5356–5364 (2019)

    work page 2019

  10. [10]

    He,K.,Zhang,X.,Ren,S.,Sun,J.:Deepresiduallearningforimagerecognition.In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 770–778 (2016)

    work page 2016

  11. [11]

    In: Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition

    He, Y.Y., Zhang, P., Wei, X.S., Zhang, X., Sun, J.: Relieving long-tailed instance segmentation via pairwise class balance. In: Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition. pp. 7000–7009 (2022)

    work page 2022

  12. [12]

    Advances in neural information processing systems33, 6840–6851 (2020)

    Ho, J., Jain, A., Abbeel, P.: Denoising diffusion probabilistic models. Advances in neural information processing systems33, 6840–6851 (2020)

    work page 2020

  13. [13]

    GPT-4o System Card

    Hurst, A., Lerer, A., Goucher, A.P., Perelman, A., Ramesh, A., Clark, A., Os- trow, A., Welihinda, A., Hayes, A., Radford, A., et al.: Gpt-4o system card. arXiv preprint arXiv:2410.21276 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  14. [14]

    In: European Conference on Computer Vision

    de Jorge, P., Volpi, R., Dokania, P.K., Torr, P.H., Rogez, G.: Placing objects in con- text via inpainting for out-of-distribution segmentation. In: European Conference on Computer Vision. pp. 456–473. Springer (2024) TMI 35

    work page 2024

  15. [15]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Kim, J., Han, S., Jeong, J., Choi, J., Kim, D., Kim, S.J.: Orida: Object-centric real-world image composition dataset. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 3051–3060 (2025)

    work page 2025

  16. [16]

    In: 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)

    Kimhi, M., Vainshtein, D., Baskin, C., Di Castro, D.: Robot instance segmentation with few annotations for grasping. In: 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV). pp. 7939–7949. IEEE (2025)

    work page 2025

  17. [17]

    Labs, B.F.: Flux.https://github.com/black-forest-labs/flux(2024)

    work page 2024

  18. [18]

    FLUX.1 Kontext: Flow Matching for In-Context Image Generation and Editing in Latent Space

    Labs, B.F., Batifol, S., Blattmann, A., Boesel, F., Consul, S., Diagne, C., Dock- horn, T., English, J., English, Z., Esser, P., et al.: Flux. 1 kontext: Flow match- ing for in-context image generation and editing in latent space. arXiv preprint arXiv:2506.15742 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  19. [19]

    In: ICLR (2024)

    Li, Y., Keuper, M., Zhang, D., Khoreva, A.: Adversarial supervision makes layout- to-image diffusion models thrive. In: ICLR (2024)

    work page 2024

  20. [20]

    Flow Matching for Generative Modeling

    Lipman, Y., Chen, R.T., Ben-Hamu, H., Nickel, M., Le, M.: Flow matching for generative modeling. arXiv preprint arXiv:2210.02747 (2022)

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  21. [21]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Liu, C., Li, X., Ding, H.: Referring image editing: Object-level image editing via referring expressions. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 13128–13138 (June 2024)

    work page 2024

  22. [22]

    Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow

    Liu, X., Gong, C., Liu, Q.: Flow straight and fast: Learning to generate and transfer data with rectified flow. arXiv preprint arXiv:2209.03003 (2022)

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  23. [23]

    Liu, Z., Lin, Y., Cao, Y., Hu, H., Wei, Y., Zhang, Z., Lin, S., Guo, B.: Swin transformer:Hierarchicalvisiontransformerusingshiftedwindows.In:Proceedings of the IEEE/CVF international conference on computer vision. pp. 10012–10022 (2021)

    work page 2021

  24. [24]

    Loiseau, T., Vu, T.H., Chen, M., Pérez, P., Cord, M.: Reliability in semantic seg- mentation: Can we use synthetic data? In: European Conference on Computer Vision. pp. 442–459. Springer (2024)

    work page 2024

  25. [25]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Luo, W., Yang, S., Zhang, X., Zhang, W.: Siedob: Semantic image editing by dis- entangling object and background. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 1868–1878 (June 2023)

    work page 2023

  26. [26]

    Lipiecki, K

    Molina, J.M., Llerena, J.P., Usero, L., Patricio, M.A.: Advances in instance seg- mentation: Technologies, metrics and applications in computer vision. Neurocom- puting625, 129584 (2025).https://doi.org/https://doi.org/10.1016/j. neucom.2025.129584,https://www.sciencedirect.com/science/article/pii/ S0925231225002565

    work page doi:10.1016/j 2025

  27. [27]

    Advances in Neural Information Processing Systems36, 76872–76892 (2023)

    Nguyen, Q., Vu, T., Tran, A., Nguyen, K.: Dataset diffusion: Diffusion-based syn- thetic data generation for pixel-level semantic segmentation. Advances in Neural Information Processing Systems36, 76872–76892 (2023)

    work page 2023

  28. [28]

    Niitani, Y., Akiba, T., Kerola, T., Ogawa, T., Sano, S., Suzuki, S.: Sampling tech- niquesforlarge-scaleobjectdetectionfromsparselyannotatedobjects.In:Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 6510–6518 (2019)

    work page 2019

  29. [29]

    Advances in Neural Information Processing Systems34, 2529–2542 (2021)

    Pan, T.Y., Zhang, C., Li, Y., Hu, H., Xuan, D., Changpinyo, S., Gong, B., Chao, W.L.: On model calibration for long-tailed object detection and instance segmen- tation. Advances in Neural Information Processing Systems34, 2529–2542 (2021)

    work page 2021

  30. [30]

    Peebles,W.,Xie,S.:Scalablediffusionmodelswithtransformers.In:Proceedingsof the IEEE/CVF international conference on computer vision. pp. 4195–4205 (2023)

    work page 2023

  31. [31]

    SAM 2: Segment Anything in Images and Videos

    Ravi, N., Gabeur, V., Hu, Y.T., Hu, R., Ryali, C., Ma, T., Khedr, H., Rädle, R., Rolland, C., Gustafson, L., et al.: Sam 2: Segment anything in images and videos. arXiv preprint arXiv:2408.00714 (2024) 36 H. Song & S. Choi et al

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  32. [32]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution image synthesis with latent diffusion models. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 10684–10695 (2022)

    work page 2022

  33. [33]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Tan, J., Lu, X., Zhang, G., Yin, C., Li, Q.: Equalization loss v2: A new gradient bal- ance approach for long-tailed object detection. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 1685–1694 (2021)

    work page 2021

  34. [34]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Tan, J., Wang, C., Li, B., Li, Q., Ouyang, W., Yin, C., Yan, J.: Equalization loss for long-tailed object recognition. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 11662–11671 (2020)

    work page 2020

  35. [35]

    Advances in neural information processing systems30(2017)

    Tarvainen, A., Valpola, H.: Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results. Advances in neural information processing systems30(2017)

    work page 2017

  36. [36]

    Tschirschwitz, D.E., Rodehorst, V.: Label convergence: Defining an upper perfor- manceboundinobjectrecognitionthroughcontradictoryannotations.In:Proceed- ings of the Winter Conference on Applications of Computer Vision. pp. 6848–6857 (2025)

    work page 2025

  37. [37]

    Classification Calibration for Long-tail Instance Segmentation

    Wang, T., Li, Y., Kang, B., Li, J., Liew, J.H., Tang, S., Hoi, S., Feng, J.: Classification calibration for long-tail instance segmentation. arXiv preprint arXiv:1910.13081 (2019)

    work page internal anchor Pith review Pith/arXiv arXiv 1910

  38. [38]

    The Devil is in Classification: A Simple Framework for Long-tail Object Detection and Instance Segmentation

    Wang, T., Li, Y., Kang, B., Li, J., Liew, J., Tang, S., Hoi, S., Feng, J.: The devil is in classification: A simple framework for long-tail instance segmentation. arXiv preprint arXiv:2007.11978 (2020)

    work page internal anchor Pith review Pith/arXiv arXiv 2007

  39. [39]

    In: Proceedings of the AAAI conference on artificial intelligence

    Wang, T., Yang, T., Cao, J., Zhang, X.: Co-mining: Self-supervised learning for sparsely annotated object detection. In: Proceedings of the AAAI conference on artificial intelligence. vol. 35, pp. 2800–2808 (2021)

    work page 2021

  40. [40]

    In: Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition

    Wang, X., Darrell, T., Rambhatla, S.S., Girdhar, R., Misra, I.: Instancediffusion: Instance-level control for image generation. In: Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition. pp. 6232–6242 (2024)

    work page 2024

  41. [41]

    Image quality assessment: From error visibility to structural similarity.Image Processing, IEEE Transactions on, 13:600 – 612, 05 2004

    Wang, Z., Bovik, A., Sheikh, H., Simoncelli, E.: Image quality assessment: from error visibility to structural similarity. IEEE Transactions on Image Processing 13(4), 600–612 (2004).https://doi.org/10.1109/TIP.2003.819861

    work page doi:10.1109/tip.2003.819861 2004

  42. [42]

    Qwen-Image Technical Report

    Wu, C., Li, J., Zhou, J., Lin, J., Gao, K., Yan, K., Yin, S.m., Bai, S., Xu, X., Chen, Y., et al.: Qwen-image technical report. arXiv preprint arXiv:2508.02324 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  43. [43]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Wu, S., Fei, H., Chua, T.s.: Universal scene graph generation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 14158–14168 (June 2025)

    work page 2025

  44. [44]

    Advances in Neural Information Processing Systems36, 54683–54695 (2023)

    Wu, W., Zhao, Y., Chen, H., Gu, Y., Zhao, R., He, Y., Zhou, H., Shou, M.Z., Shen, C.: Datasetdm: Synthesizing data with perception annotations using diffu- sion models. Advances in Neural Information Processing Systems36, 54683–54695 (2023)

    work page 2023

  45. [45]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Wu, W., Zhao, Y., Shou, M.Z., Zhou, H., Shen, C.: Diffumask: Synthesizing images with pixel-level annotations for semantic segmentation using diffusion models. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 1206–1217 (2023)

    work page 2023

  46. [46]

    In: Conference on Robot Learning

    Xie, C., Mousavian, A., Xiang, Y., Fox, D.: Rice: Refining instance masks in clut- tered environments with graph neural networks. In: Conference on Robot Learning. pp. 1655–1665. PMLR (2022)

    work page 2022

  47. [47]

    International Journal of Computer Vision133(4), 1456–1475 (2025) TMI 37

    Xie, J., Li, W., Li, X., Liu, Z., Ong, Y.S., Loy, C.C.: Mosaicfusion: Diffusion mod- els as data augmenters for large vocabulary instance segmentation. International Journal of Computer Vision133(4), 1456–1475 (2025) TMI 37

    work page 2025

  48. [48]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Xue,H.,Huang,Z.,Sun,Q.,Song,L.,Zhang,W.:Freestylelayout-to-imagesynthe- sis. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 14256–14266 (2023)

    work page 2023

  49. [49]

    Advances in Neural Infor- mation Processing Systems36, 18659–18675 (2023)

    Yang, L., Xu, X., Kang, B., Shi, Y., Zhao, H.: Freemask: Synthetic images with dense annotations make stronger segmentation models. Advances in Neural Infor- mation Processing Systems36, 18659–18675 (2023)

    work page 2023

  50. [50]

    In: European Conference on Computer Vision

    Ye, H., Kuen, J., Liu, Q., Lin, Z., Price, B., Xu, D.: Seggen: Supercharging segmen- tation models with text2mask and mask2img synthesis. In: European Conference on Computer Vision. pp. 352–370. Springer (2024)

    work page 2024

  51. [51]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Zhang, L., Rao, A., Agrawala, M.: Adding conditional control to text-to-image diffusion models. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 3836–3847 (2023)

    work page 2023

  52. [52]

    IEEE Transactions on Robotics (2025)

    Zhang, Y., Yin, M., Bi, W., Yan, H., Bian, S., Zhang, C.H., Hua, C.: Zisvfm: Zero-shot object instance segmentation in indoor robotic environments with vision foundation models. IEEE Transactions on Robotics (2025)

    work page 2025

  53. [53]

    In: International Conference on Machine Learning

    Zhao, H., Sheng, D., Bao, J., Chen, D., Chen, D., Wen, F., Yuan, L., Liu, C., Zhou, W., Chu, Q., et al.: X-paste: Revisiting scalable copy-paste for instance seg- mentation using clip and stablediffusion. In: International Conference on Machine Learning. pp. 42098–42109. PMLR (2023)

    work page 2023

  54. [54]

    doi:10.1109/ICASSP49357.2023.10095642 , abstract =

    Zhao, Y., Chen, S., Chen, Q., Hu, Z.: Combining loss reweighting and sample resampling for long-tailed instance segmentation. In: ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). pp. 1–5 (2023).https://doi.org/10.1109/ICASSP49357.2023.10094303

    work page doi:10.1109/icassp49357.2023.10094303 2023

  55. [55]

    Probabilistic two-stage detection

    Zhou, X., Koltun, V., Krähenbühl, P.: Probabilistic two-stage detection. arXiv preprint arXiv:2103.07461 (2021)

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  56. [56]

    InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models

    Zhu, J., Wang, W., Chen, Z., Liu, Z., Ye, S., Gu, L., Tian, H., Duan, Y., Su, W., Shao, J., et al.: Internvl3: Exploring advanced training and test-time recipes for open-source multimodal models. arXiv preprint arXiv:2504.10479 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  57. [57]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Zong, Z., Song, G., Liu, Y.: Detrs with collaborative hybrid assignments training. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 6748–6758 (2023)

    work page 2023