Pith. sign in

REVIEW 4 major objections 6 minor 70 references

Document Image Rectification Bases on Self-Adaptive Multitask Fusion

T0 review · 4 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read The paper claims that document rectification reaches state-of-the-art accuracy when four auxiliary tasks—3D coordinates, UV maps, and horizontal/vertical lines—are fused through a leave-one-out feature aggregation module and a gating…

desk verdict The gating and aggregation modules are credible and the self-ablation holds up, but the unqualified three-benchmark SOTA claim is contradicted by the paper's own DocReal table, and the training-data mismatch keeps the headline result conditional. read the letter →

arxiv 2505.06038 v1 pith:TSBMN3VJ submitted 2025-05-09 cs.CV

classification cs.CV
keywords documentimagerectificationdewarpingmultitasklearningfeatureaggregationgatingmechanismtextlinesegmentation3DcoordinatepredictionUVmap
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 tries to establish that document image rectification improves when auxiliary tasks are fused self-adaptively rather than concatenated. It proposes SalmRec, which predicts 3D coordinates, UV maps, and horizontal/vertical lines, then uses an inter-task feature aggregation module and a gating mechanism to combine them into a 2D deformation field. On DIR300, DocUNet, and DocReal, it reports the best published numbers on several metrics, including the lowest character error rate on DIR300. The significance is that task interference, not lack of auxiliary information, may be the bottleneck.

What carries the argument

The load-bearing mechanism is the inter-task feature aggregation (FA) module paired with a gating module. FA concatenates three of the four task features, derives a global attention vector from them, multiplies it with the remaining task's feature, and adds a residual connection; this lets each task borrow complementary geometry while suppressing redundant information. The gating module computes channel-wise softmax weights from the shared encoder feature and applies them separately to the global pair (3D/UV) and the local pair (horizontal/vertical lines), before both are concatenated and fed to a transformer decoder that predicts the 2D deformation grid. A weighted line loss supervises the line tasks, and a grid-sampling decoder produces the final rectified image.

What would settle it

Re-run each published baseline checkpoint through the same OCR engine and metric implementation used for Tables 1-3 while retraining that baseline on the same DocDewarpHV1 training set; if the reported margins disappear or reverse, the state-of-the-art claim collapses.

Watch

Extended reading notes

Core claim

The paper's central claim is that the bottleneck in multi-task document rectification is the way auxiliary tasks are combined, not the auxiliary information itself. SalmRec predicts four complementary signals—3D coordinates, UV texture coordinates, horizontal text lines, and vertical lines—and merges them with an inter-task feature aggregation module that uses a leave-one-out global-attention design, followed by a gating mechanism that weighs global features (3D/UV) and local features (lines) separately. With this combination, the paper reports a state-of-the-art result on the DIR300 benchmark (MS-SSIM 0.67, LD 5.14, AD 0.178, ED 444.07, CER 0.1400), best geometric metrics among compared methods on DocReal, and the best recognition metrics among compared methods in the 60-image OCR setting on DocUNet. The paper reads this as evidence that negative interference between tasks can be reduced by adaptive fusion.

Load-bearing premise

The headline comparison assumes that re-running published baselines through the authors' own evaluation pipeline and training on a different annotated synthetic dataset yields numbers comparable to originally published results.

Editorial extensions

If this is right

  • Auxiliary supervision is not automatically beneficial: the paper shows that all four tasks together without fusion underperform the fused version on DIR300 and DocReal, so future multi-task dewarpers need explicit interaction modeling.
  • The reported gains put the best OCR accuracy on DIR300 (CER 0.14) ahead of prior methods that also use text-line attention, suggesting that fused local constraints transfer to recognition.
  • If the numbers hold, SalmRec becomes the reference point on DIR300 and DocReal, with the fusion and gating modules as the components to ablate in follow-up work.
  • The architecture's shared encoder and two-gate design imply that additional auxiliary tasks, such as layout masks, could be absorbed by adding one more task branch and one more gate.

Reading between the lines

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

  • The paper trains on DocDewarpHV1 rather than the widely used Doc3D set, so the reported margins mix the effect of the fusion modules with the effect of better line annotations; an ablation that trains the same network and the leading baselines on Doc3D would isolate the modules' true contribution.
  • Because the failure cases in the paper's own limitation section involve strong shadows, heavy folds, and backgrounds that match the paper edge, the method's practical reach is likely bounded by the same conditions that limit its auxiliary-task predictions.
  • The leave-one-out aggregation idea is exportable: any set of complementary geometric tasks, such as depth, normal, and edge prediction in generic image restoration, could use the same residual global-attention gating rather than naive concatenation.
  • A testable compression experiment would replace the 7.1M-parameter gating module with a per-pixel routing layer and check whether the reported gains survive; if they do, the mechanism, not its parameter count, is what matters.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The manuscript proposes SalmRec, a multitask document image rectification network that predicts a 2D deformation field with four auxiliary tasks: 3D coordinate maps, UV maps, horizontal lines, and vertical lines. The network includes an inter-task feature aggregation module (FA) that reweights each task's features using a leave-one-out global attention mechanism, and a gating module (Gate) that balances global and local task features before a Transformer decoder predicts the deformation field. The model is trained on the synthetic DocDewarpHV1 dataset and evaluated on three benchmarks: DIR300, DocReal, and DocUNet. The paper reports state-of-the-art results on all three benchmarks and presents ablations showing that the full model with FA and Gate outperforms the four-task baseline without these modules.

Significance. The paper addresses a real and current problem in document dewarping: negative interference among auxiliary tasks in multitask rectification networks. The proposed FA and Gate modules are simple, interpretable, and the authors' self-contained ablation (Tables 4 and 5) consistently shows that the full model improves over the four-task baseline on all three benchmarks, giving internal support for the architecture. If the claimed results hold under controlled comparison, the work would be a useful contribution to the document image rectification literature. However, the external significance is substantially weakened by two issues: the headline state-of-the-art claim is internally contradicted by the authors' own Table 2 on DocReal, and the main comparisons are confounded by training on a different synthetic dataset (DocDewarpHV1) than most baselines (Doc3D). The paper does not provide code or trained models, and the absence of a controlled retraining experiment makes the cross-dataset comparison difficult to interpret.

major comments (4)
  1. [§1 and §4.4 (Table 2)] The unqualified claim in the introduction and abstract that the method 'achieves state-of-the-art performance on three public benchmarks' is contradicted by Table 2 on DocReal. In that table, Ours reports ED 193.06 and CER 0.2604, while PaperEdge reports ED 186.86 and CER 0.2451, and DocReal reports ED 184.54 and CER 0.2485. On two of the five reported metrics, Ours is not the best. The text in §4.4 correctly limits the improvement over DocReal to MS-SSIM, LD, and AD, but the abstract and contribution bullet emphasize text recognition accuracy as part of the state-of-the-art claim. This internal inconsistency must be resolved, either by qualifying the SOTA claim to geometric metrics or by providing a reasoned justification for why the OCR results on DocReal should not be taken at face value.
  2. [§4.1 and Tables 1–3] The state-of-the-art comparisons are confounded by a training-data mismatch. The model is trained on DocDewarpHV1 (Section 4.1), while most of the compared baselines were trained on Doc3D. Different synthetic training data can change rectification performance substantially, especially for OCR metrics, so the reported improvements may partly reflect the training distribution rather than the proposed architecture. The manuscript notes that Doc3D lacks fine-grained annotations and that prior methods use binarization or line-detection models, which is a legitimate motivation for using DocDewarpHV1. However, the authors then compare their numbers to published results from models trained on Doc3D without a controlled experiment. To support the state-of-the-art claim, the authors should retrain, or otherwise evaluate, baselines on the same training set, or carefully discuss the expected effect of the training-data difference and provide controlled ablations to separate the contribution of the architecture from the contribution of the training data.
  3. [§4.5, Table 5] The ablation of the individual FA and Gate modules in Table 5 shows mixed and sometimes negative effects on the DocUNet benchmark: adding FA alone (row 2) worsens LD from 7.28 to 7.30 and ED from 350.5 to 376.8 relative to the four-task baseline, and adding Gate alone (row 3) worsens MS-SSIM from 0.50 to 0.49 and LD from 7.28 to 7.49. The full model (row 4) improves over the baseline on all three benchmarks, and the paper states that the two modules jointly serve the task. This is a defensible position, but the current presentation does not discuss why a module that alone can degrade performance is still beneficial when combined. The authors should address this apparent negative interference explicitly, for example by analyzing the learned gate weights or by reporting per-benchmark variance, to strengthen the claim that the combination is robust and not a result of fortuitous hyperparameter choices.
  4. [§4.3, OCR evaluation protocol] The OCR evaluation uses two different engines (Tesseract for English, PaddleOCR for Chinese) and different image subsets for the two English benchmarks (90 images for DIR300, '50 and 60 images' for DocUNet). The sentence in Section 4.3 saying 'the text recognition accuracy is evaluated on 50 and 60 images, respectively' is ambiguous when read with Table 3, which shows two ED/CER numbers per method. It appears that the first number in each pair corresponds to a 50-image subset and the second to a 60-image subset, but this is not stated explicitly in the table or its caption. Because the SOTA claim depends on the comparability of these numbers, the authors must clarify the exact evaluation protocol and ensure that all methods are evaluated on the same image subsets with the same OCR engine and preprocessing.
minor comments (6)
  1. [Title] The word 'Bases' in the title should be 'Based'.
  2. [Figure 6 caption] The caption reads 'Predcit results of our proposed method'; 'Predcit' should be 'Prediction'.
  3. [Table 3] The method name 'Pice-Wise' should be 'Piece-Wise' in both the table and the text in Section 4.4.
  4. [Section 4.5] In the paragraph discussing Table 5, the sentence 'The Character Error Rate (CER) reached an impressive 10.3%.' is unclear because it is not stated whether 10.3% is a relative or absolute improvement. Please specify the reference baseline.
  5. [Section 4.2] The sentence 'On the DocUNet benchmark, the text recognition accuracy is evaluated on 50 and 60 images, respectively' could be interpreted as two separate evaluations; please clarify the exact subsets and whether the results are reported as pairs in Table 3.
  6. [Section 4.5] The word 'benchamrks' appears in the ablation discussion; it should be 'benchmarks'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is an empirical architecture study whose benchmark claims are evaluated on independent test sets, and its only self-citation is a baseline comparison, not a load-bearing premise.

full rationale

SalmRec is presented as an end-to-end multi-task rectification network, and its central claims are empirical: the network is trained on the synthetic DocDewarpHV1 dataset with supervised auxiliary tasks (3D coordinates, UV maps, horizontal/vertical lines) and evaluated on the public DIR300, DocUNet, and DocReal benchmarks. The loss functions in Eqs. (3)-(7) are standard supervised losses on ground-truth annotations and grid coordinates; no quantity in the loss is defined in terms of the reported benchmark metrics, so the results are not forced by construction. The auxiliary task labels come from the training dataset, not from the test benchmarks, and the ablation study in Tables 4 and 5 varies modules and tasks while holding the evaluation protocol fixed, so the reported improvements are not fitted to the test sets. The only self-citation is FTDR [25], a previous work by overlapping authors; it is used as a related method and as a comparison baseline, but the proposed architecture does not depend on FTDR's results or code for its validity, so this self-citation is not load-bearing. The table note that DocReal baselines were 'evaluated by ourself from the public code or rectified results' raises a legitimate comparability/correctness concern, and the paper's own Table 2 shows that Ours is not best on DocReal ED/CER, contradicting the unqualified 'state-of-the-art' claim; however, these are issues of empirical validity and internal consistency, not circularity. The Limitations section acknowledges failure cases without attempting to redefine them away. No equation reduces to its own input, no fitted parameter is renamed as a prediction, and no uniqueness theorem or prior-work ansatz is used to force the architecture choice. The paper is therefore not circular, though its headline SOTA statement is vulnerable on DocReal.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical entities. It relies on standard deep learning assumptions about synthetic-to-real transfer and metric validity, plus the central modeling premise that the four auxiliary tasks are complementary. The free parameters listed are hand-set hyperparameters that affect the reported results but are not derived from theory.

free parameters (3)
  • lambda (loss balance weight) = 5
    Chosen to balance rectification and segmentation losses (Eq. 7); no sensitivity analysis is provided.
  • Decoder layer loss weight factor 1/(2L-i) = 1/(2L-i) for layer i, L=4
    Hand-defined weighting of auxiliary losses across decoder depths (Eq. 5).
  • Number of transformer attention layers L_a = 6
    Architecture choice; no ablation is reported.
assumptions (4)
  • domain assumption Models trained on synthetic distorted document images (DocDewarpHV1) generalize to real-world benchmarks (DIR300, DocReal, DocUNet).
    Standard assumption in the field; the paper provides no analysis of the synthetic-to-real gap.
  • standard math The evaluation metrics MS-SSIM, LD, AD, ED, and CER are meaningful proxies for rectification quality.
    These metrics are widely used in prior work; not independently justified here.
  • domain assumption The ground-truth 3D coordinates, UV maps, and line annotations in DocDewarpHV1 are accurate and unbiased.
    Training supervision depends on the correctness of the synthetic dataset; no validation of annotation quality is presented.
  • ad hoc to paper Auxiliary tasks (3D, UV, lines) provide complementary information that helps predict the 2D deformation field.
    This is the paper's core modeling premise; supported only by the ablation tables, not by an external principle.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Document Image Rectification Bases on Self-Adaptive Multitask Fusion." pith.science (2026). https://pith.science/paper/TSBMN3VJ

@misc{pith2026250506038,
  author       = {Pith},
  title        = {Pith review of: Document Image Rectification Bases on Self-Adaptive Multitask Fusion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TSBMN3VJ}},
  note         = {Machine review of arXiv:2505.06038}
}
read the original abstract

Deformed document image rectification is essential for real-world document understanding tasks, such as layout analysis and text recognition. However, current multi-task methods -- such as background removal, 3D coordinate prediction, and text line segmentation -- often overlook the complementary features between tasks and their interactions. To address this gap, we propose a self-adaptive learnable multi-task fusion rectification network named SalmRec. This network incorporates an inter-task feature aggregation module that adaptively improves the perception of geometric distortions, enhances feature complementarity, and reduces negative interference. We also introduce a gating mechanism to balance features both within global tasks and between local tasks effectively. Experimental results on two English benchmarks (DIR300 and DocUNet) and one Chinese benchmark (DocReal) demonstrate that our method significantly improves rectification performance. Ablation studies further highlight the positive impact of different tasks on dewarping and the effectiveness of our proposed module.

Figures

Figures reproduced from arXiv: 2505.06038 by the authors.

Figure 1
Figure 1. Overview architecture of our proposed SalmRec. For a given distorted image, SalmRec learns to predict a 2D deformation field by multi-decoder segmentation module (𝐷𝑖 ), inter-task feature aggregation (𝐹 𝐴), gating mechanisms (𝐺𝑖 ), and Transformer decoder. 𝑭 represents the output feature map. global features, some methods [15, 17, 66, 16, 48, 24] remove background as a preprocessing stage. Often a lightweight networ… view at source ↗
Figure 2
Figure 2. Inter-task Feature Aggregation (FA) Module. The four tasks are divided into four groups, where 𝑭3𝐷 is fused with the global attention score obtained from the other three tasks for feature interconnection aggregation. (BN) [21] and Rectified Linear Unit (ReLU) activation [42] blocks (ConvBNReLU, CBR) are used to extract the general representation of the input image 𝑰𝒓 in first layer. The remaining four layers are dow… view at source ↗
Figure 3
Figure 3. Gating Mechanism. 𝑭𝑎 is the universal feature extracted from the original distorted image by the encoder of the segmentation module. 𝑭 ′ 3𝐷 and 𝑭 ′ 𝑢𝑣 are two aggregated features with global properties. Similar operations are performed on local features 𝑭 ′ ℎ𝑙𝑖𝑛𝑒 and 𝑭 ′ 𝑣𝑙𝑖𝑛𝑒. 𝐶, 𝐻, 𝑊 represents the channel, height and width of the feature map respectively. Transformer Decoder. After the global and local elements a… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Qualitative visual comparison with existing methods on DIR300 benchmark [17]. The first column “Distorted” means the given distorted document image, and the last column “Ground-Truth” is the flattened reference image. 4.4. Experimental Results Performance on DIR300 Ben…
Figure 5
Figure 5. Figure 5: Qualitative visual comparison with existing methods on DocReal benchmark [64]. deformed images of different types and scenes, like folded and indoor (first two rows of [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: Predcit results of our proposed method. The first column “Distorted” is the given distorted document image and the last column “Rectified” represents the image rectified by our model. These three samples are from DocUNet [40], DIR300 [17] and DocReal [64] benchmark res…
Figure 7
Figure 7. Figure 7: A few images are used for sample analysis. Sample 1 and 2 are sampled from DIR300 [17] and DocReal [64] benchmark. Sample 1 image (first row) has strong folding except for shadows. Sample 2 (second row) has grid-shaped texture interference in the background, and the ba…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

70 extracted references · 53 canonical work pages

  1. [1]

    Srrv: A novel document object detector based on spatial-related relation and vision

    Bi, H., Xu, C., Shi, C., Liu, G., Li, Y., Zhang, H., Qu, J., 2023. Srrv: A novel document object detector based on spatial-related relation and vision. IEEE Transactions on Multimedia 25, 3788–3798

  2. [2]

    Multitask feature learning as multiobjective optimization: A new genetic programming approach to image classification

    Bi, Y., Xue, B., Zhang, M., 2022. Multitask feature learning as multiobjective optimization: A new genetic programming approach to image classification. IEEE Transactions on Cybernetics 53, 3007–3020

  3. [3]

    A tutorial on the cross-entropy method

    de Boer, P.T., Kroese, D.P., Mannor, S., Rubinstein, R.Y., 2005. A tutorial on the cross-entropy method. Annals of Operations Research 134, 19–67

  4. [4]

    Geometric and shading correction for images of printed materials using boundary

    Brown, M.S., Tsoi, Y.C., 2006. Geometric and shading correction for images of printed materials using boundary. IEEE Transactions on Image Processing 15, 1544–1554

  5. [6]

    Instructir: High-quality image restoration following human instructions

    Conde, M.V., Geigle, G., Timofte, R., 2024. Instructir: High-quality image restoration following human instructions

  6. [7]

    Shape from shading for the digitization of curved documents

    Courteille, F., Crouzil, A., Durou, J.D., Gurdjos, P., 2007. Shape from shading for the digitization of curved documents. Machine Vision and Applications 18, 301–316

  7. [8]

    Matadoc: Margin and text aware document dewarping for arbitrary boundary

    Dai, B., Li, X., Xie, Q., Li, Y., Qin, X., Zhang, C., Yao, K., Han, J., 2023. Matadoc: Margin and text aware document dewarping for arbitrary boundary. ArXiv abs/2307.12571

  8. [9]

    Dewarpnet: Single-image document unwarping with stacked 3d and 2d regression networks

    Das, S., Ma, K., Shu, Z., Samaras, D., Shilkrot, R., 2019. Dewarpnet: Single-image document unwarping with stacked 3d and 2d regression networks. 2019 IEEE/CVF International Conference on Computer Vision (ICCV) , 131–140

Show all 70 references
  1. [10]

    End-to-endpiece-wiseunwarpingofdocumentimages

    Das,S.,Singh,K.Y.,Wu,J.,Bas,E.,Mahadevan,V.,Bhotika,R.,Samaras,D.,2021. End-to-endpiece-wiseunwarpingofdocumentimages. 2021 IEEE/CVF International Conference on Computer Vision (ICCV) , 4248–4257

  2. [11]

    Autosegedge: Searching for the edge device real-time semantic segmentation based on multi-task learning

    Dou, Z., Ye, D., Wang, B., 2023. Autosegedge: Searching for the edge device real-time semantic segmentation based on multi-task learning. Image and Vision Computing 136, 104719. doi:https://doi.org/10.1016/j.imavis.2023.104719. Heng Li et al. Page 12 of 14 Document Image Recti...

  3. [12]

    Svtr: Scene text recognition with a single visual model, in: International Joint Conference on Artificial Intelligence

    Du, Y., Chen, Z., Jia, C., Yin, X., Zheng, T., Li, C., Du, Y., Jiang, Y.G., 2022. Svtr: Scene text recognition with a single visual model, in: International Joint Conference on Artificial Intelligence

  4. [13]

    Docpedia: Unleashing the power of large multimodal model in the frequency domain for versatile document understanding

    Feng, H., Liu, Q., Liu, H., gang Zhou, W., Li, H., Huang, C., 2023a. Docpedia: Unleashing the power of large multimodal model in the frequency domain for versatile document understanding. ArXiv abs/2311.11810

  5. [14]

    Deep unrestricted document image rectification

    Feng, H., Liu, S., Deng, J., Zhou, W., Li, H., 2024. Deep unrestricted document image rectification. IEEE Transactions on Multimedia 26, 6142–6154. doi:10.1109/TMM.2023.3347094

  6. [15]

    Doctr: Document image transformer for geometric unwarping and illumination correction

    Feng, H., Wang, Y., gang Zhou, W., Deng, J., Li, H., 2021a. Doctr: Document image transformer for geometric unwarping and illumination correction. Proceedings of the 29th ACM International Conference on Multimedia

  7. [16]

    Docscanner: Robust document image rectification with progressive learning

    Feng, H., gang Zhou, W., Deng, J., Tian, Q., Li, H., 2021b. Docscanner: Robust document image rectification with progressive learning. ArXiv abs/2110.14968

  8. [17]

    Geometric representation learning for document image rectification, in: European Conference on Computer Vision

    Feng, H., gang Zhou, W., Deng, J., Wang, Y., Li, H., 2022. Geometric representation learning for document image rectification, in: European Conference on Computer Vision

  9. [18]

    Multi-task probabilistic regression with overlap maximization for visual tracking

    Feng, Z., Yan, L., Xia, Y., Xiao, B., 2023b. Multi-task probabilistic regression with overlap maximization for visual tracking. IEEE Transactions on Circuits and Systems for Video Technology 33, 7554–7564

  10. [19]

    Iimt-net: Poly-1 weights balanced multi-task network for semantic segmentation and depth estimation using interactive information

    He, M., Yang, Z., Zhang, G., Long, Y., Song, H., 2024. Iimt-net: Poly-1 weights balanced multi-task network for semantic segmentation and depth estimation using interactive information. Image and Vision Computing 148, 105109. doi:https://doi.org/10.1016/j.imavis. 2024.105109

  11. [20]

    Text line extraction of curved document images using hybrid metric

    Huang, Z., Gu, J., Meng, G., Pan, C., 2015. Text line extraction of curved document images using hybrid metric. 2015 3rd IAPR Asian Conference on Pattern Recognition (ACPR) , 251–255

  12. [21]

    Batch normalization: Accelerating deep network training by reducing internal covariate shift

    Ioffe, S., Szegedy, C., 2015. Batch normalization: Accelerating deep network training by reducing internal covariate shift. ArXiv abs/1502.03167

  13. [22]

    Revisitingdocumentimagedewarpingbygridregularization

    Jiang,X.,Long,R.,Xue,N.,Yang,Z.,Yao,C.,Xia,G.,2022. Revisitingdocumentimagedewarpingbygridregularization. 2022IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , 4533–4542

  14. [23]

    Segment anything

    Kirillov, A., Mintun, E., Ravi, N., Mao, H., Rolland, C., Gustafson, L., Xiao, T., Whitehead, S., Berg, A.C., Lo, W.Y., Dollár, P., Girshick, R.B., 2023. Segment anything. 2023 IEEE/CVF International Conference on Computer Vision (ICCV) , 3992–4003

  15. [24]

    Amireadable?transferlearningbaseddocumentimagerectification

    Kumari,P.,Das,S.,2024. Amireadable?transferlearningbaseddocumentimagerectification. Int.J.DocumentAnal.Recognit.27,433–446

  16. [25]

    Foreground and text-lines aware document image rectification, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp

    Li, H., Wu, X., Chen, Q., Xiang, Q., 2023a. Foreground and text-lines aware document image rectification, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 19574–19583

  17. [26]

    Volter:Visualcollaborationanddual-streamfusionforscenetextrecognition

    Li,J.N.,Liu,X.Q.,Luo,X.,Xu,X.S.,2024. Volter:Visualcollaborationanddual-streamfusionforscenetextrecognition. IEEETransactions on Multimedia 26, 6437–6448

  18. [27]

    Layout-aware single-image document flattening

    Li, P., Quan, W., Guo, J., Yan, D.M., 2023b. Layout-aware single-image document flattening. ACM Transactions on Graphics 43, 1–17

  19. [28]

    Documentrectificationandilluminationcorrectionusingapatch-basedcnn

    Li,X.,Zhang,B.,Liao,J.,Sander,P.V.,2019. Documentrectificationandilluminationcorrectionusingapatch-basedcnn. ACMTransactions on Graphics (TOG) 38, 1 – 11

  20. [29]

    Geometricrectificationofcamera-captureddocumentimages

    Liang,J.,DeMenthon,D.,Doermann,D.S.,2008. Geometricrectificationofcamera-captureddocumentimages. IEEETransactionsonPattern Analysis and Machine Intelligence 30, 591–605

  21. [30]

    Real-time scene text detection with differentiable binarization and adaptive scale fusion

    Liao, M., Zou, Z., Wan, Z., Yao, C., Bai, X., 2023. Real-time scene text detection with differentiable binarization and adaptive scale fusion. IEEE Transactions on Pattern Analysis and Machine Intelligence 45, 919–931. doi:10.1109/TPAMI.2022.3155612

  22. [31]

    Doclayllm: An efficient and effective multi-modal extension of large language models for text-rich document understanding

    Liao, W., Wang, J., Li, H., Wang, C., Huang, J., Jin, L., 2024. Doclayllm: An efficient and effective multi-modal extension of large language models for text-rich document understanding. ArXiv abs/2408.15045

  23. [32]

    Sift flow: Dense correspondence across scenes and its applications

    Liu, C., Yuen, J., Torralba, A., 2011. Sift flow: Dense correspondence across scenes and its applications. IEEE Transactions on Pattern Analysis and Machine Intelligence 33, 978–994

  24. [33]

    Docmae: Document image rectification via self-supervised representation learning

    Liu, S., Feng, H., gang Zhou, W., Li, H., Liu, C., Wu, F., 2023. Docmae: Document image rectification via self-supervised representation learning. 2023 IEEE International Conference on Multimedia and Expo (ICME) , 1613–1618

  25. [34]

    Geometricrectificationofdocumentimagesusingadversarialgatedunwarpingnetwork

    Liu,X.,Meng,G.,Fan,B.,Xiang,S.,Pan,C.,2020. Geometricrectificationofdocumentimagesusingadversarialgatedunwarpingnetwork. Pattern Recognit. 108, 107576

  26. [35]

    Textadapter:Self-superviseddomainadaptationforcross-domaintextrecognition

    Liu,X.Q.,Zhang,P.,Luo,X.,Huang,Z.H.,Xu,X.S.,2024. Textadapter:Self-superviseddomainadaptationforcross-domaintextrecognition. IEEE Transactions on Multimedia 26, 9854–9865

  27. [36]

    Global attention mechanism: Retain information to enhance channel-spatial interactions

    Liu, Y., Shao, Z., Hoffmann, N., 2021. Global attention mechanism: Retain information to enhance channel-spatial interactions. ArXiv abs/2112.05561

  28. [37]

    Decoupled weight decay regularization, in: International Conference on Learning Representations

    Loshchilov, I., Hutter, F., 2017. Decoupled weight decay regularization, in: International Conference on Learning Representations

  29. [38]

    Excavation reinforcement learning using geometric representation

    Lu, Q., Zhu, Y., Zhang, L., 2022. Excavation reinforcement learning using geometric representation. IEEE Robotics and Automation Letters PP, 1–1

  30. [39]

    Learningfromdocumentsinthewildtoimprovedocumentunwarping

    Ma,K.,Das,S.,Shu,Z.,Samaras,D.,2022. Learningfromdocumentsinthewildtoimprovedocumentunwarping. ACMSIGGRAPH2022 Conference Proceedings

  31. [40]

    Docunet:Documentimageunwarpingviaastackedu-net

    Ma,K.,Shu,Z.,Bai,X.,Wang,J.,Samaras,D.,2018. Docunet:Documentimageunwarpingviaastackedu-net. 2018IEEE/CVFConference on Computer Vision and Pattern Recognition , 4700–4709

  32. [41]

    Exploiting vector fields for geometric rectification of distorted document images, in: European Conference on Computer Vision

    Meng, G., Su, Y., Wu, Y., Xiang, S., Pan, C., 2018. Exploiting vector fields for geometric rectification of distorted document images, in: European Conference on Computer Vision

  33. [42]

    Rectifiedlinearunitsimproverestrictedboltzmannmachines,in:InternationalConferenceonMachineLearning

    Nair,V.,Hinton,G.E.,2010. Rectifiedlinearunitsimproverestrictedboltzmannmachines,in:InternationalConferenceonMachineLearning

  34. [43]

    Automatic differentiation in pytorch

    Paszke, A., Gross, S., Chintala, S., Chanan, G., Yang, E., DeVito, Z., Lin, Z., Desmaison, A., Antiga, L., Lerer, A., 2017. Automatic differentiation in pytorch

  35. [44]

    Geometric multimodal contrastive representation learning, in: International Conference on Machine Learning

    Poklukar, P., Vasco, M., Yin, H., Melo, F.S., Paiva, A., Kragic, D., 2022. Geometric multimodal contrastive representation learning, in: International Conference on Machine Learning. Heng Li et al. Page 13 of 14 Document Image Rectification Bases on Self-Adaptive Multitask Fusion

  36. [45]

    Ronneberger,O.,Fischer,P.,Brox,T.,2015. U-net:Convolutionalnetworksforbiomedicalimagesegmentation,in:Medicalimagecomputing andcomputer-assistedintervention–MICCAI2015:18thinternationalconference,Munich,Germany,October5-9,2015,proceedings,partIII 18, Springer. pp. 234–241

  37. [46]

    An overview of the tesseract ocr engine

    Smith, R., 2007. An overview of the tesseract ocr engine. Ninth International Conference on Document Analysis and Recognition (ICDAR

  38. [47]

    Strezoski,G.,vanNoord,N.,Worring,M.,2019.Manytasklearningwithtaskrouting.2019IEEE/CVFInternationalConferenceonComputer Vision (ICCV) , 1375–1384

  39. [48]

    Efficientjointrectificationofphotometricandgeometricdistortionsindocumentimages

    Tang,H.,Guo,J.,Wang,T.,Yu,Y.,Wang,C.,2024. Efficientjointrectificationofphotometricandgeometricdistortionsindocumentimages. ICASSP 2024 - 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , 3690–3694

  40. [49]

    Multi-view document rectification using boundary

    Tsoi, Y.C., Brown, M.S., 2007. Multi-view document rectification using boundary. 2007 IEEE Conference on Computer Vision and Pattern Recognition , 1–8

  41. [50]

    UVDoc: Neural grid-based document unwarping, in: SIGGRAPH ASIA, Technical Papers

    Verhoeven, F., Magne, T., Sorkine-Hornung, O., 2023. UVDoc: Neural grid-based document unwarping, in: SIGGRAPH ASIA, Technical Papers. https://doi.org/10.1145/3610548.3618174

  42. [51]

    Geometric representation learning

    Vilnis, L., 2021. Geometric representation learning

  43. [52]

    Shape from shading with interreflections under a proximal light source: Distortion-free copying of an unfolded book

    Wada, T., Ukida, H., Matsuyama, T., 1997. Shape from shading with interreflections under a proximal light source: Distortion-free copying of an unfolded book. International Journal of Computer Vision 24, 125–135

  44. [53]

    Deep learning image segmentation based on adaptive total variation preprocessing

    Wang, G., Ma, Y., Pan, Z., Zhang, X., 2024. Deep learning image segmentation based on adaptive total variation preprocessing. IEEE Transactions on Cybernetics 54, 7991–7998

  45. [54]

    Domainadaptationmultitaskoptimization

    Wang,X.,Kang,Q.,Zhou,M.,Yao,S.,Abusorrah,A.M.,2022. Domainadaptationmultitaskoptimization. IEEETransactionsonCybernetics 53, 4567–4578

  46. [55]

    Image quality assessment: from error visibility to structural similarity

    Wang, Z., Bovik, A.C., Sheikh, H.R., Simoncelli, E.P., 2004. Image quality assessment: from error visibility to structural similarity. IEEE Transactions on Image Processing 13, 600–612

  47. [56]

    General ocr theory: Towards ocr-2.0 via a unified end-to-end model

    Wei, H., Liu, C., Chen, J., Wang, J., Kong, L., Xu, Y., Ge, Z., Zhao, L., Sun, J., Peng, Y., Han, C., Zhang, X., 2024. General ocr theory: Towards ocr-2.0 via a unified end-to-end model. ArXiv abs/2409.01704

  48. [57]

    Tabledetectionforvisuallyrichdocumentimages

    Xiao,B.,Simsek,M.,Kantarci,B.,Alkheir,A.A.,2023. Tabledetectionforvisuallyrichdocumentimages. Knowl.BasedSyst.282,111080

  49. [58]

    Dewarping document image by displacement flow estimation with fully convolutional network

    Xie, G.W., Yin, F., Zhang, X.Y., Liu, C.L., 2020. Dewarping document image by displacement flow estimation with fully convolutional network. ArXiv abs/2104.06815

  50. [59]

    Documentdewarpingwithcontrolpoints,in:IEEEInternationalConferenceonDocument Analysis and Recognition

    Xie,G.W.,Yin,F.,Zhang,X.Y.,Liu,C.L.,2022a. Documentdewarpingwithcontrolpoints,in:IEEEInternationalConferenceonDocument Analysis and Recognition

  51. [60]

    C2 am: Contrastive learning of class-agnostic activation map for weakly supervisedobjectlocalizationandsemanticsegmentation

    Xie, J., Xiang, J., Chen, J., Hou, X., Zhao, X., Shen, L., 2022b. C2 am: Contrastive learning of class-agnostic activation map for weakly supervisedobjectlocalizationandsemanticsegmentation. 2022IEEE/CVFConferenceonComputerVisionandPatternRecognition(CVPR) , 979–988

  52. [61]

    Xue,C.,Tian,Z.,Zhan,F.,Lu,S.,Bai,S.,2022.Fourierdocumentrestorationforrobustdocumentdewarpingandrecognition.2022IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , 4563–4572

  53. [62]

    Dan: Distortion-aware network for fisheye image rectification using graph reasoning

    Yan, Y., Liu, H., Zhang, C., Xu, C., Xu, B., Pan, W., Dai, S., Song, Y., 2025. Dan: Distortion-aware network for fisheye image rectification using graph reasoning. Image and Vision Computing 156, 105423. doi:https://doi.org/10.1016/j.imavis.2025.105423

  54. [63]

    Multiview rectification of folded documents

    You, S., Matsushita, Y., Sinha, S., Bou, Y.K., Ikeuchi, K., 2016. Multiview rectification of folded documents. IEEE Transactions on Pattern Analysis and Machine Intelligence 40, 505–511

  55. [64]

    Yu,F.,Xie,Y.,Wu,L.,Wen,Y.,Wang,G.,Ren,S.,Chen,X.,Mao,J.,Li,W.,2024. Docreal:Robustdocumentdewarpingofreal-lifeimages via attention-enhanced control point prediction, in: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pp. 665–674

  56. [65]

    Cpfusion: A multi-focus image fusion method based on closed-loop regularization

    Zhai, H., Chen, P., Luo, N., Li, Q., Yu, P., 2025. Cpfusion: A multi-focus image fusion method based on closed-loop regularization. Image and Vision Computing 154, 105399. doi:https://doi.org/10.1016/j.imavis.2024.105399

  57. [66]

    Marior: Margin removal and iterative content rectification for document dewarping in the wild

    Zhang, J., Luo, C., Jin, L., Guo, F., Ding, K., 2022a. Marior: Margin removal and iterative content rectification for document dewarping in the wild. Proceedings of the 30th ACM International Conference on Multimedia

  58. [67]

    Docres: A generalist model toward unifying document image restoration tasks

    Zhang, J., Peng, D., Liu, C., Zhang, P., Jin, L., 2024a. Docres: A generalist model toward unifying document image restoration tasks. 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , 15654–15664

  59. [68]

    Dockylin: A large multimodal model for visual document understanding with efficient visual slimming

    Zhang, J., Yang, W., Lai, S., Xie, Z., Jin, L., 2024b. Dockylin: A large multimodal model for visual document understanding with efficient visual slimming. ArXiv abs/2406.19101

  60. [69]

    Multimodalpre-trainingbasedongraphattentionnetworkfordocumentunderstanding

    Zhang,Z.,Ma,J.,Du,J.,Wang,L.,Zhang,J.,2022b. Multimodalpre-trainingbasedongraphattentionnetworkfordocumentunderstanding. IEEE Transactions on Multimedia 25, 6743–6755

  61. [70]

    Enhancing geometric factors in modellearning and inference forobject detection and instance segmentation

    Zheng, Z., Wang,P., Ren, D., Liu,W., Ye, R., Hu, Q.,Zuo, W., 2020. Enhancing geometric factors in modellearning and inference forobject detection and instance segmentation. IEEE Transactions on Cybernetics 52, 8574–8586

  62. [71]

    Multi-taskdisagreement-reducingmultimodalsentimentfusionnetwork

    Zijun,W.,Naicheng,J.,Xinyue,C.,Bin,S.,2024. Multi-taskdisagreement-reducingmultimodalsentimentfusionnetwork. ImageandVision Computing 149, 105158. doi:https://doi.org/10.1016/j.imavis.2024.105158. Heng Li et al. Page 14 of 14

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.