pith. sign in

arxiv: 2606.31446 · v1 · pith:6LSTZUJTnew · submitted 2026-06-30 · 💻 cs.CL · cs.CV

Revising RVL-CDIP: Quantifying Errors and Test-Train Overlap

Stefan Larson , Attila Nagy , Sam Desai , Cyrus Desai , Nicole C. Lima , Yixin Yuan , Siddharth Betala , Kaushal K. Prajapati
show 3 more authors
Jamiu T. Suleiman Sharad Duwal Kevin Leach
This is my paper

Pith reviewed 2026-07-01 05:49 UTC · model grok-4.3

classification 💻 cs.CL cs.CV
keywords RVL-CDIPdocument classificationlabel errorstest-train overlapdata qualityout-of-distribution generalization
0
0 comments X

The pith

RVL-CDIP contains 12% label errors and 35% test-train duplication, and correcting label errors improves out-of-distribution accuracy by an average of 8.1 percentage points.

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

The paper audits the RVL-CDIP dataset for document classification and identifies substantial label errors and test-train overlap. It quantifies these issues at 12% label error and 35% duplication between train and test splits. By producing corrected dataset variants, the work shows that training on error-corrected data leads to better performance on a separate out-of-distribution benchmark. This indicates that standard reported accuracies on the original dataset may be inflated by data quality problems.

Core claim

Our rigorous analysis of RVL-CDIP finds that the corpus contains 12% label error and approximately 35% test-train duplication. Remediation sees improvements in classification accuracy when errors are removed, but sees decreases in accuracy when duplicates are removed. We additionally evaluate models on RVL-CDIP-N, an out-of-distribution benchmark, finding that training on error-corrected data substantially improves OOD generalization, with supervised models gaining an average of 8.1 percentage points in accuracy and improvements as large as 14 percentage points.

What carries the argument

Systematic detection of label errors and test-train overlaps via automated and manual procedures, followed by creation of revised dataset variants for benchmarking.

If this is right

  • Training on error-corrected data improves OOD accuracy by an average of 8.1 percentage points.
  • Removing duplicates from the training set decreases accuracy on the original test set.
  • Supervised models show gains as large as 14 percentage points on the out-of-distribution benchmark after label correction.

Where Pith is reading between the lines

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

  • Other document classification benchmarks may contain comparable rates of label errors and leakage that affect reported results.
  • Dataset creators should apply similar detection steps as a standard part of release.
  • Some previously published performance numbers on RVL-CDIP may partly reflect exploitation of errors and duplicates rather than true generalization.

Load-bearing premise

The automated and manual procedures used to detect label errors and test-train overlaps correctly identify true errors and duplicates at the reported rates without substantial false positives that would alter the 12% and 35% figures.

What would settle it

An independent review of a sample of flagged items finding that a substantial fraction of the identified label errors are actually correct would undermine the 12% error rate and the reported accuracy gains from correction.

Figures

Figures reproduced from arXiv: 2606.31446 by Attila Nagy, Cyrus Desai, Jamiu T. Suleiman, Kaushal K. Prajapati, Kevin Leach, Nicole C. Lima, Sam Desai, Sharad Duwal, Siddharth Betala, Stefan Larson, Yixin Yuan.

Figure 1
Figure 1. Figure 1: Example label errors (top row) and test-train near [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Example documents from a selection of RVL-CDIP’s 16 categories. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Examples of RVL-CDIP label errors with no valid true label. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Examples from RVL-CDIP with multiple valid labels. Top labels: original; bottom labels: alternate label. [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Examples of incorrectly labeled samples from RVL-CDIP with a known correct label. [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Example duplicate (top) and near-duplicate (bottom) [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Example (near-) duplicate test-train pairs. Documents need not be pixel-identical to constitute leakage; template-match [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
read the original abstract

RVL-CDIP is a popular dataset for benchmarking document classifiers. However, the dataset contains ample amounts of label errors as well as non-trivial amounts of test-train overlap, both of which may impact model performance metrics. In this paper, we address these two problems by (1) finding and fixing label errors, and (2) detecting and addressing test-train overlap. We produce several variations of RVL-CDIP with label error and test-train overlap fixes, and benchmark document classification performance on these new RVL-CDIP variations. Our rigorous analysis of RVL-CDIP finds that the corpus contains 12\% label error and approximately 35% test-train duplication. Remediation sees improvements in classification accuracy when errors are removed, but sees decreases in accuracy when duplicates are removed. We additionally evaluate models on RVL-CDIP-N, an out-of-distribution benchmark, finding that training on error-corrected data substantially improves OOD generalization, with supervised models gaining an average of 8.1 percentage points in accuracy and improvements as large as 14 percentage points.

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 analyzes the RVL-CDIP document classification benchmark and reports that it contains 12% label errors and approximately 35% test-train duplication. The authors produce corrected dataset variants, benchmark supervised models on them, and evaluate out-of-distribution generalization on RVL-CDIP-N, claiming average accuracy gains of 8.1 percentage points (up to 14 pp) when training on error-corrected data.

Significance. If the detection pipeline is reliable, the work is significant because it quantifies previously unmeasured data-quality problems in a widely used benchmark and shows measurable effects on both in-distribution and OOD performance. Releasing the corrected dataset variants is a concrete contribution that enables follow-up work.

major comments (2)
  1. [Methods] Methods section (label-error detection pipeline): No precision, recall, or inter-annotator agreement figures are reported for the combined automated+manual procedure that produces the 12% label-error rate. Because the headline percentages and the 8.1 pp OOD gains rest directly on the correctness of these detections, absence of validation metrics leaves the central empirical claims without an independent check on false-positive rate.
  2. [Results] Results on RVL-CDIP-N (OOD evaluation): The reported accuracy deltas are obtained by training on the authors' corrected training sets; if the label-error detector has non-negligible false positives, the 'corrected' sets contain spurious changes and the measured generalization improvement cannot be attributed solely to error removal.
minor comments (2)
  1. [Abstract] Abstract and §4: the exact definition of 'test-train duplication' (exact image match, near-duplicate, or label-consistent duplicate) should be stated explicitly when the 35% figure is first introduced.
  2. [Results] Table 2 or equivalent: clarify whether the accuracy numbers are macro-averaged or micro-averaged and whether the same train/validation split protocol is used across all corrected variants.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for their careful reading and constructive feedback on the label-error detection pipeline and its implications for the OOD results. We address each major comment below and indicate planned revisions.

read point-by-point responses

  1. Referee: [Methods] Methods section (label-error detection pipeline): No precision, recall, or inter-annotator agreement figures are reported for the combined automated+manual procedure that produces the 12% label-error rate. Because the headline percentages and the 8.1 pp OOD gains rest directly on the correctness of these detections, absence of validation metrics leaves the central empirical claims without an independent check on false-positive rate.

    Authors: We agree that the absence of precision, recall, and inter-annotator agreement metrics is a limitation. The pipeline used automated candidate generation based on embedding similarity followed by manual review and correction performed by the authors. These metrics were not computed because the automated step served only to surface candidates rather than as an evaluated classifier, and manual review was conducted by a single team without multiple annotators. We will revise the Methods section to describe the manual review process in greater detail, state the lack of these quantitative validation figures explicitly, and note this as a limitation. The consistent OOD gains across models provide indirect support for the overall utility of the corrections. revision: partial

  2. Referee: [Results] Results on RVL-CDIP-N (OOD evaluation): The reported accuracy deltas are obtained by training on the authors' corrected training sets; if the label-error detector has non-negligible false positives, the 'corrected' sets contain spurious changes and the measured generalization improvement cannot be attributed solely to error removal.

    Authors: This concern follows directly from the previous point and is valid. Without validation metrics on the detector, it is possible that some corrections were erroneous and that part of the observed 8.1 pp average (up to 14 pp) OOD improvement could arise from other factors. We will add a paragraph in the Results and Discussion sections acknowledging this possibility, noting that the multi-model consistency and the out-of-distribution nature of RVL-CDIP-N reduce the likelihood that gains are due solely to in-distribution artifacts, and reiterating the limitation. The reported numbers themselves will remain unchanged. revision: partial

standing simulated objections not resolved
  • Quantitative precision, recall, or inter-annotator agreement figures for the label-error detection pipeline, as these were not computed in the original study.

Circularity Check

0 steps flagged

No circularity: empirical measurement of dataset errors

full rationale

The paper conducts an empirical analysis by applying automated and manual procedures to detect and quantify label errors (reported at 12%) and test-train overlaps (reported at ~35%) directly in the RVL-CDIP corpus, then measures accuracy changes on the resulting revised datasets. No mathematical derivation, fitted parameters, or self-citation chain reduces these quantities to quantities defined by the authors' own choices; the central claims rest on direct data inspection rather than any of the enumerated circular patterns. The accuracy deltas (e.g., 8.1 pp OOD gain) are likewise measured outcomes on the corrected data, not predictions forced by construction. This is a standard self-contained empirical measurement study.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claims rest on the assumption that the authors' label-error and duplicate-detection procedures are sufficiently accurate; no free parameters or invented entities are described in the abstract.

axioms (2)
  • domain assumption Label error detection procedure identifies true errors at the scale needed to support the 12% figure
    Directly supports the headline quantification and all downstream accuracy claims
  • domain assumption Test-train overlap detection procedure identifies true duplicates at the scale needed to support the 35% figure
    Directly supports the headline quantification and the deduplication ablation results

pith-pipeline@v0.9.1-grok · 5757 in / 1342 out tokens · 27107 ms · 2026-07-01T05:49:08.998945+00:00 · methodology

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

56 extracted references · 10 canonical work pages · 6 internal anchors

  1. [1]

    Cathaoir Agnew, Ciarán Eising, Patrick Denny, Anthony Scanlan, Pepijn Van De Ven, and Eoin M. Grua. 2023. Quantifying the Effects of Ground Truth Annotation Quality on Object Detection and Instance Segmentation Performance. IEEE Access11 (2023), 25174–25188

    2023

  2. [2]

    Miltiadis Allamanis. 2019. The adverse effects of code duplication in machine learning models of code. InProceedings of the 2019 ACM SIGPLAN International Symposium on New Ideas, New Paradigms, and Reflections on Programming and Software

    2019

  3. [3]

    Souhail Bakkali, Zuheng Ming, Mickaël Coustaty, and Marçal Rusiñol. 2021. EAML: ensemble self-attention-based mutual learning network for document image classification.Int. J. Doc. Anal. Recognit.24, 3 (2021), 251–268

    2021

  4. [4]

    Björn Barz and Joachim Denzler. 2020. Do We Train on Test Data? Purging CIFAR of Near-Duplicates.Journal of Imaging6, 6 (2020)

    2020

  5. [5]

    Longformer: The Long-Document Transformer

    Iz Beltagy, Matthew E. Peters, and Arman Cohan. 2020. Longformer: The Long- Document Transformer.arXiv:2004.05150(2020)

    work page internal anchor Pith review Pith/arXiv arXiv 2020

  6. [6]

    Ali Babar, and M

    Roland Croft, M. Ali Babar, and M. Mehdi Kholoosi. 2023. Data Quality for Soft- ware Vulnerability Datasets. InProceedings of the 45th International Conference on Software Engineering (ICSE)

    2023

  7. [7]

    Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabinovich. 2018. SuperPoint: Self-Supervised Interest Point Detection and Description. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

    2018

  8. [8]

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT)

    2019

  9. [9]

    Masato Fujitake. 2024. LayoutLLM: Large Language Model Instruction Tuning for Visually Rich Document Understanding. InProceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)

    2024

  10. [10]

    Navarini, and Marc Pouly

    Fabian Gröger, Simone Lionetti, Philippe Gottfrois, Alvaro Gonzalez-Jimenez, Matthew Groh, Roxana Daneshjou, Labelling Consortium, Alexander A. Navarini, and Marc Pouly. 2023. Towards Reliable Dermatology Evaluation Benchmarks. InProceedings of the 3rd Machine Learning for Health Symposium (Proceedings of Machine Learning Research, Vol. 225). PMLR, 101–128

    2023

  11. [11]

    Harley, Alex Ufkes, and Konstantinos G

    Adam W. Harley, Alex Ufkes, and Konstantinos G. Derpanis. 2015. Evaluation of Deep Convolutional Nets for Document Image Classification and Retrieval. In Proceedings of the International Conference on Document Analysis and Recognition (ICDAR). DocEng ’26, August 25–28, 2026, Fribourg, Switzerland Larson et al

    2015

  12. [12]

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep Residual Learning for Image Recognition. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

    2016

  13. [13]

    Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, and Furu Wei. 2022. LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking. InProceed- ings of the 30th ACM International Conference on Multimedia

    2022

  14. [14]

    Le, Yunhsuan Sung, Zhen Li, and Tom Duerig

    Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc V. Le, Yunhsuan Sung, Zhen Li, and Tom Duerig. 2021. Scaling Up Visual and Vision- Language Representation Learning With Noisy Text Supervision. InProceedings of the 38th International Conference on Machine Learning (ICML)

    2021

  15. [15]

    Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, De- vendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, Lélio Renard Lavaud, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroix, and William El Sayed. 2023. Mistral 7B.arXiv preprint ...

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  16. [16]

    Michael Jungo, Lars Vögtlin, Atefeh Fakhari, Nathan Wegmann, Rolf Ingold, Andreas Fischer, and Anna Scius-Bertrand. 2023. Impact of the ground truth quality for handwriting recognition. InProceedings of the 12th International Symposium on Information and Communication Technology

    2023

  17. [17]

    Geewook Kim, Teakgyu Hong, Moonbin Yim, JeongYeon Nam, Jinyoung Park, Jinyeong Yim, Wonseok Hwang, Sangdoo Yun, Dongyoon Han, and Seunghyun Park. 2022. OCR-Free Document Understanding Transformer. InProceedings of the European Conference on Computer Vision (ECCV)

    2022

  18. [18]

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton. 2012. ImageNet Classi- fication with Deep Convolutional Neural Networks. InProceedings of the 25th International Conference on Neural Information Processing Systems

    2012

  19. [19]

    Seif Laatiri, Pirashanth Ratnamogan, Joël Tang, Laurent Lam, William Vanhuffel, and Fabien Caspani. 2023. Information Redundancy and Biases in Public Doc- ument Information Extraction Benchmarks. InProceedings of the International Conference on Document Analysis and Recognition (ICDAR)

    2023

  20. [20]

    Richard Landis and Gary G

    J. Richard Landis and Gary G. Koch. 1977. The Measurement of Observer Agree- ment for Categorical Data.Biometrics33, 1 (1977), 159–174

    1977

  21. [21]

    Britto, Rodrigo Minetto, and David Menotti

    Rayson Laroca, Valter Estevam, Alceu S. Britto, Rodrigo Minetto, and David Menotti. 2023. Do We Train on Test Data? The Impact of Near-Duplicates on License Plate Recognition. InProceedings of the International Joint Conference on Neural Networks (IJCNN)

    2023

  22. [22]

    Stefan Larson, Gordon Lim, Yutong Ai, David Kuang, and Kevin Leach. 2022. Evaluating Out-of-Distribution Performance on Document Image Classifiers. In Advances in Neural Information Processing Systems

    2022

  23. [23]

    Stefan Larson, Gordon Lim, and Kevin Leach. 2023. On Evaluation of Docu- ment Classification with RVL-CDIP. InProceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics (EACL)

    2023

  24. [24]

    Patrick Lewis, Pontus Stenetorp, and Sebastian Riedel. 2021. Question and Answer Test-Train Overlap in Open-Domain Question Answering Datasets. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics (EACL)

    2021

  25. [25]

    Boda Li, Gabriele Baris, Pak Hung Chan, Anima Rahman, and Valentina Donzella

  26. [26]

    InProceedings of the 2022 International Conference on Electrical, Computer, Communications and Mechatronics Engineering (ICECCME)

    Testing ground-truth errors in an automotive dataset for a DNN-based object detector. InProceedings of the 2022 International Conference on Electrical, Computer, Communications and Mechatronics Engineering (ICECCME). 1–6

    2022

  27. [27]

    Junlong Li, Yiheng Xu, Tengchao Lv, Lei Cui, Cha Zhang, and Furu Wei. 2022. DiT: Self-supervised Pre-training for Document Image Transformer. InProceedings of the 30th ACM International Conference on Multimedia

    2022

  28. [28]

    Gordon Lim, Stefan Larson, and Kevin Leach. 2024. Label Errors in the To- bacco3482 Dataset.arXiv preprint arXiv:2412.13140(2024)

    work page arXiv 2024

  29. [29]

    Philipp Lindenberger, Paul-Edouard Sarlin, and Marc Pollefeys. 2023. Light- Glue: Local Feature Matching at Light Speed. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)

    2023

  30. [30]

    Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. RoBERTa: A Robustly Optimized BERT Pretraining Approach.arXiv preprint arXiv:1907.11692 (2019)

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  31. [31]

    Chuwei Luo, Guozhi Tang, Qi Zheng, Cong Yao, Lianwen Jin, Chenliang Li, Yang Xue, and Luo Si. 2025. Bi-VLDoc: bidirectional vision-language modeling for visually-rich document understanding.International Journal on Document Analysis and Recognition (IJDAR)(2025)

    2025

  32. [32]

    Meta AI. 2025. The Llama 4 Herd: The Beginning of a New Era of Natively Multimodal AI Innovation

    2025

  33. [33]

    Yida Mu, Mali Jin, Xingyi Song, and Nikolaos Aletras. 2024. Enhancing Data Qual- ity through Simple De-duplication: Navigating Responsible Computational Social Science Research. InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing (EMNLP)

    2024

  34. [34]

    Müller and Karla Markert

    Nicolas M. Müller and Karla Markert. 2019. Identifying Mislabeled Instances in Classification Datasets. InProceedings of the International Joint Conference on Neural Networks (IJCNN)

    2019

  35. [35]

    Jingcheng Niu and Gerald Penn. 2019. Rationally Reappraising ATIS-based Dialogue Systems. InProceedings of the 57th Annual Meeting of the Association for Computational Linguistics (ACL)

    2019

  36. [36]

    Northcutt, Anish Athalye, and Jonas Mueller

    Curtis G. Northcutt, Anish Athalye, and Jonas Mueller. 2021. Pervasive Label Errors in Test Sets Destabilize Machine Learning Benchmarks. InProceedings of the 35th Conference on Neural Information Processing Systems Track on Datasets and Benchmarks

    2021

  37. [37]

    OpenAI. 2025. gpt-oss-120b & gpt-oss-20b Model Card. arXiv:2508.10925

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  38. [38]

    Filip Radenovic, Ahmet Iscen, Giorgos Tolias, Yannis Avrithis, and Ondrej Chum

  39. [39]

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

    Revisiting Oxford and Paris: Large-Scale Image Retrieval Benchmarking. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 5706–5715

  40. [40]

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. 2021. Learning Transferable Visual Models From Natural Language Supervision. InProceedings of the 38th International Conference on Machine Learning (ICML)

    2021

  41. [41]

    Ratcliff and David E

    John W. Ratcliff and David E. Metzener. 1988. Pattern Matching: The Gestalt Approach.Dr. Dobb’s Journal(1988)

    1988

  42. [42]

    Susanna Rücker and Alan Akbik. 2023. CleanCoNLL: A Nearly Noise-Free Named Entity Recognition Dataset. InProceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP)

    2023

  43. [43]

    Saifullah Saifullah, Stefan Agne, Andreas Dengel, and Sheraz Ahmed. 2024. DocX- classifier: towards a robust and interpretable deep neural network for document image classification.International Journal on Document Analysis and Recognition (IJDAR)(2024)

    2024

  44. [44]

    Anna Scius-Bertrand, Michael Jungo, Lars Vögtlin, Jean-Marc Spat, and Andreas Fischer. 2024. Zero-Shot Prompting and Few-Shot Fine-Tuning: Revisiting Docu- ment Image Classification Using Large Language Models. InProceedings of the 27th International Conference on Pattern Recognition (ICPR)

    2024

  45. [45]

    Karen Simonyan and Andrew Zisserman. 2015. Very Deep Convolutional Net- works for Large-Scale Image Recognition. InProceedings of the International Conference on Learning Representations (ICLR)

    2015

  46. [46]

    Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, and Andrew Rabinovich. 2015. Going deeper with convolutions. InProceedings of the IEEE Conference on Com- puter Vision and Pattern Recognition (CVPR)

    2015

  47. [47]

    Zineng Tang, Ziyi Yang, Guoxin Wang, Yuwei Fang, Yang Liu, Chenguang Zhu, Michael Zeng, Cha Zhang, and Mohit Bansal. 2023. Unifying Vision, Text, and Layout for Universal Document Processing. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition (CVPR)

    2023

  48. [48]

    Michael Tschannen, Alexey Gritsenko, Xiao Wang, Muhammad Ferjad Naeem, Ibrahim Alabdulmohsin, Nikhil Parthasarathy, Talfan Evans, Lucas Beyer, Ye Xia, Basil Mustafa, Olivier Hénaff, Jeremiah Harmsen, Andreas Steiner, and Xiaohua Zhai. 2025. SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Featur...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  49. [49]

    Vu and Diep Thi-Ngoc Nguyen

    Hieu M. Vu and Diep Thi-Ngoc Nguyen. 2020. Revising FUNSD Dataset for Key- Value Detection in Document Images.arXiv preprint arXiv:2010.05322(2020)

    work page arXiv 2020

  50. [50]

    Wei-Chen Wang and Jonas Mueller. 2019. Detecting Label Errors in Token Classification Data.arXiv preprint arXiv:2210.03920(2019)

    work page arXiv 2019

  51. [51]

    Haiyu Wu, Sicong Tian, Jacob Gutierrez, Aman Bhatta, Kağan Öztürk, and Kevin W. Bowyer. 2024. Identity Overlap Between Face Recognition Train/Test Data: Causing Optimistic Bias in Accuracy Measurement.arXiv preprint arXiv:2405.09403(2024)

    work page arXiv 2024

  52. [52]

    Saining Xie, Ross Girshick, Piotr Dollár, Zhuowen Tu, and Kaiming He. 2017. Aggregated Residual Transformations for Deep Neural Networks. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

    2017

  53. [53]

    Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, and Ming Zhou. 2020. LayoutLM: Pre-training of Text and Layout for Document Image Understanding. InProceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining

    2020

  54. [54]

    An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, Chujie Zheng, Dayiheng Liu, Fan Zhou, Fei Huang, Feng Hu, Hao Ge, Haoran Wei, Huan Lin, Jialong Tang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jing Zhou, Jingren Zhou, Junyang Lin, Kai Dang, Keqin Bao, Kexin Yang, ...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  55. [55]

    Manzil Zaheer, Guru Guruganesh, Kumar Avinava Dubey, Joshua Ainslie, Chris Alberti, Santiago Ontanon, Philip Pham, Anirudh Ravula, Qifan Wang, Li Yang, et al. 2020. Big bird: Transformers for longer sequences.Advances in Neural Information Processing Systems33 (2020)

    2020

  56. [56]

    Xiaohua Zhai, Basil Mustafa, Alexander Kolesnikov, and Lucas Beyer. 2023. Sig- moid Loss for Language Image Pre-Training. InProceedings of the 2023 IEEE/CVF International Conference on Computer Vision (ICCV)

    2023