Pith. sign in

REVIEW 4 major objections 6 minor 51 references

A real-captured blur benchmark that kills the low-gradient shortcut lets models detect local motion blur better with only 633 images.

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

2026-07-14 11:48 UTC pith:F3GXJNUS

load-bearing objection Real multi-camera blur-mask set that transfers with 633 images, plus a DCT-UNet that wins in-domain; hardness story is correlational but the tables are useful. the 4 major comments →

arxiv 2607.10427 v1 pith:F3GXJNUS submitted 2026-07-11 cs.CV

BOCCHI: A More Realistic and Challenging Benchmark for Local Motion Blur Detection with MSDCT-UNet

Kuan-Lin Chen , Yuan-Kang Lee , Cheng-Yuan Chiang , Jian-Jiun Ding This is my paper
classification cs.CV
keywords motion blur detectionlocal motion blur datasetDCT attentionfrequency-domain learningsemantic segmentationBOCCHIMSDCT-UNet
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.

Local motion blur detection means marking, at every pixel, which parts of a photo are smeared by moving objects while the rest stays sharp. Prior benchmarks often let models cheat by treating low-gradient patches as blur; those shortcuts collapse when the same model faces new cameras or scenes. This paper builds BOCCHI, 633 real photographs from five consumer cameras whose sharp regions deliberately share the same gradient range as the blurred ones, forcing models to use frequency evidence instead. It also introduces MSDCT-UNet, which feeds multi-scale DCT features into every stage of a U-Net via attention and FiLM fusion. On BOCCHI the new network leads in mask accuracy and boundary quality; more importantly, any model trained only on BOCCHI transfers better to a held-out mixed test set than models trained on larger existing sources.

Core claim

BOCCHI’s sharp-region gradients overlap the blur-region gradient distribution more than any prior local-motion-blur dataset, defeating the low-gradient-equals-blur shortcut; models trained on its 633 images therefore achieve the strongest cross-dataset transfer, and the frequency-aware MSDCT-UNet that injects multi-scale DCT priors ranks first in-domain on BOCCHI.

What carries the argument

MSDCT-UNet: a U-Net that, at every encoder and decoder stage, extracts 57-channel multi-scale high-frequency DCT features from the Sobel gradient map, reweights them with multi-head DCT Attention, and fuses them into spatial features via FiLM modulation plus an attentive frequency ASPP bottleneck.

Load-bearing premise

The claim that BOCCHI’s transfer gains come mainly from its sharp–blur gradient overlap, rather than from multi-camera diversity, inclusive annotation style, or the makeup of the mixed test set.

What would settle it

Retrain the same models on a version of BOCCHI whose sharp-region gradients have been filtered to remove the low-gradient tail (so PR25 no longer overlaps blur) and check whether the cross-dataset mIoU advantage over ReLoBlur and OMoBlur disappears.

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

4 major / 6 minor

Summary. The paper introduces BOCCHI, a 633-image real multi-camera benchmark for pixel-level local motion blur detection, arguing that its sharp-region gradient distribution overlaps the blur distribution (high PR25/μ_blur) and thereby defeats low-gradient=blur shortcuts present in prior datasets. It also proposes MSDCT-UNet, a U-Net-style encoder–decoder that injects multi-scale DCT features via multi-head DCT Attention, FiLM fusion, and an AFASPP bottleneck. Empirically, MSDCT-UNet ranks first in-domain on BOCCHI in mIoU and BdF1 among 13 models, and models trained on BOCCHI achieve the highest average cross-dataset mIoU/Dice/Recall on a 572-image mixed Inference Dataset despite fewer training images than ReLoBlur or OMoBlur.

Significance. If the empirical transfer advantage and hardness characterization hold under tighter controls, BOCCHI would be a useful community resource for local blur detection, where existing sets are small (CUHKmotion), scene-biased (ReLoBlur), or soft-mask synthetic (OMoBlur). The multi-camera real capture, human polygon annotations, and explicit gradient-overlap statistics are concrete contributions. The architecture is a solid engineering synthesis of HiFST-style DCT priors with modern fusion (FiLM, SE, ASPP) rather than a foundational theoretical advance, but the frequency-transform ablation and architecture ablations give a clear inductive-bias story. The architecture-agnostic transfer pattern (13/13 models) is a strength of the experimental design if the evaluation set is not biased toward BOCCHI’s capture protocol.

major comments (4)
  1. Abstract / Sec. 3.3 / Table 3: The central causal claim—that BOCCHI’s superior cross-dataset transfer is because sharp–blur gradient overlap (PR25/μ_blur ≈ 0.68) defeats gradient shortcuts—is not isolated from confounds. BOCCHI also differs in multi-camera PSF diversity, inclusive motion-artifact annotation style, and scene mix. The Inference Dataset is author-constructed with 28.7% BOCCHI-protocol images (Sec. 3.3). Without per-subset transfer tables (BOCCHI-protocol vs ReLoBlur vs OMoBlur vs CUHK held-out) and/or a controlled comparison that varies only the gradient-overlap statistic while holding annotation protocol and camera population fixed, the “defeats shortcuts” explanation remains correlational. Soften the causal language or add the subset breakdowns and a confound discussion as primary results.
  2. Table 2–3 and Sec. 5.1–5.2: All headline numbers are single-run (seed 531). The in-domain margin over Cellpose3 is +1.2 pp mIoU / +4.0 pp BdF1, and the average cross-dataset edge over OMoBlur is only +2.2 pp mIoU. The paired 38/63 win rate with +0.7 pp median mIoU is suggestive but not a substitute for multi-seed means and standard deviations (or bootstrap CIs). For claims that BOCCHI “outperforms every other training source” and that MSDCT-UNet “ranks first,” report at least 3 seeds for the main BOCCHI and cross-dataset averages, or qualify rankings as single-run.
  3. Sec. 5.3 / Table 2: The hardness argument rests heavily on DDRNet-23 trailing MSDCT-UNet by 14.9 pp on BOCCHI vs 2.0 pp on ReLoBlur. That comparison mixes architecture capacity and frequency bias. To support “BOCCHI more strongly penalizes gradient shortcuts,” also report (i) a pure spatial ablation of MSDCT-UNet (already in Table 4 as w/o Freq Branch) trained and tested on each source, and (ii) the same spatial-only backbone’s cross-dataset transfer when trained on BOCCHI vs ReLoBlur. Currently the 7.5× amplification is informative but incomplete as a benchmark-hardness proof.
  4. Sec. 3.2 Annotation: Three candidate polygons with majority selection is good practice, but no inter-annotator agreement (e.g., mean pairwise IoU or boundary disagreement) is reported. Because annotation inclusiveness (“all visible motion artifacts”) is itself a plausible driver of transfer differences versus ReLoBlur/OMoBlur soft masks, IAA statistics and a short protocol comparison are load-bearing for interpreting Table 3.
minor comments (6)
  1. Supplementary Table 7: MSDCT-UNet at 1.0 FPS (1080×720) is a severe practical limitation relative to STDC/BiSeNet-class baselines. Mention this constraint in the main-text conclusion or limitations, not only the supplement.
  2. Eq. (4): The 0.1·tanh(γ) residual scale is a free hyperparameter; briefly justify the 0.1 choice or note it was not tuned extensively.
  3. Fig. 3 / Table 1: Mask convention “black→blur, white→sharp” is stated in the table caption but inverted in some qualitative figure descriptions (Fig. 8 says black = predicted blur). Harmonize mask polarity across all figures.
  4. Sec. 4.2: Channel count 57 is inherited from HiFST; a short sensitivity check (e.g., 32 vs 57 vs 96) in the supplement would strengthen the design choice beyond “design constant.”
  5. Related Work: OMoBlur is cited as contemporaneous CVPR 2026; ensure the camera-ready citation and any concurrent-work disclaimer match the final venue timeline.
  6. Typos / polish: “CUHK motion” vs “CUHKmotion” inconsistency; “AF ASPP” spacing in Fig. 1; occasional missing spaces after periods in the abstract PDF text.

Circularity Check

0 steps flagged

No circularity: empirical CV benchmark + architecture paper; metrics and transfer results are not forced by construction from inputs or self-citation.

full rationale

The paper is a standard empirical computer-vision contribution: it releases a real-captured, human-annotated dataset (BOCCHI), proposes an encoder-decoder that injects multi-scale DCT features (inspired by the external HiFST prior of Golestaneh & Karam 2017, not by the present authors), and reports ordinary segmentation metrics (mIoU, Dice, BdF1) under identical training protocols against 12 external baselines and three external training sources. The strongest claims (in-domain ranking on BOCCHI; superior average cross-dataset transfer of BOCCHI-trained models on the held-out Inference set) are measured quantities, not algebraic identities or fitted parameters renamed as predictions. DCT channel count (57), temperature au=1.2, deep-supervision weights, and FiLM scale factor 0.1 are ordinary hyper-parameters; they do not define the reported scores. No uniqueness theorem, self-citation load-bearing premise, or self-definitional relation appears. Confounds in causal interpretation of why BOCCHI transfers well (annotation style, multi-camera PSF, Inference-set composition) are validity concerns, not circularity. The derivation chain is therefore self-contained and non-circular.

Axiom & Free-Parameter Ledger

6 free parameters · 4 axioms · 3 invented entities

Load-bearing content is empirical: a new labeled dataset, classical HiFST-style DCT blur priors treated as domain knowledge, and many hand-chosen architectural hyperparameters. No formal derivation; claims rest on measured mIoU/transfer under those design choices and on the assumption that gradient-overlap statistics index shortcut hardness.

free parameters (6)
  • DCT channel count (57 smallest HF magnitudes)
    Inherited from HiFST as a multi-scale design constant; not derived from BOCCHI data but fixed by hand and central to the frequency branch.
  • DCT Attention temperature τ=1.2 and K=4 heads
    Hand-chosen softmax temperature and head count that shape frequency weighting; affect training dynamics and reported accuracy.
  • FiLM residual scale factor 0.1 and tanh-bounded γ
    Ad-hoc modulation strength chosen so fusion starts near identity; free design parameter of the fusion path.
  • Deep supervision weights α∈{1.0,0.4,0.2,0.1}
    Hand-set loss weights on main and auxiliary heads; standard practice but free relative to the data.
  • Positive BCE class weight (batch-estimated, clip 50)
    Data-dependent but clipped hyperparameter that rebalances blur vs sharp pixels each batch.
  • Training resolution 270×180, 60 epochs, lr 1e-4, seed 531
    Optimization and split choices that determine the reported numbers; single-seed results depend on them.
axioms (4)
  • domain assumption Motion blur systematically attenuates high-frequency content so that the weakest multi-scale HF DCT magnitudes on gradient maps are a stable local blur indicator (HiFST prior).
    Sec. 4.2 and supplementary B treat this classical prior as given when selecting the 57 smallest HF coefficients.
  • domain assumption Pixel-level human polygon annotations (majority of three candidates) correctly define ground-truth local motion blur, including all visible motion artifacts.
    Sec. 3.2; all supervised metrics and transfer claims rest on these labels.
  • ad hoc to paper Sharp-region PR25 below blur-region mean gradient (high PR25/μ_blur) implies models cannot rely on low-gradient=blur shortcuts and must use frequency cues.
    Sec. 3.3 / Fig. 3; this is the paper’s operational definition of benchmark hardness and is not independently proven to be the cause of transfer gains.
  • domain assumption Binary semantic segmentation with BCE+Dice and standard encoder-decoder inductive biases is an appropriate problem formulation for local motion blur detection.
    Sec. 1 and 4.1; shared with the baseline suite.
invented entities (3)
  • BOCCHI dataset no independent evidence
    purpose: Provide real multi-camera, human-annotated local motion blur masks with sharp–blur gradient overlap to defeat gradient shortcuts.
    New labeled corpus introduced by the authors; independent evidence would require public release and third-party use, not shown in the text.
  • MSDCT-UNet (DCT Attention + FiLM fusion + AFASPP) no independent evidence
    purpose: Inject multi-scale DCT blur priors at every encoder/decoder stage for pixel blur masks.
    New architecture assembly; validated only by the paper’s own ablations and tables, not by external replications.
  • Inference Dataset (572 mixed held-out images) no independent evidence
    purpose: Balanced cross-dataset transfer testbed assembled from BOCCHI-protocol, ReLoBlur, CUHKmotion, and OMoBlur subsets.
    Author-constructed evaluation mix; composition choices can favor or disfavor training sources.

pith-pipeline@v1.1.0-grok45 · 35792 in / 3683 out tokens · 46164 ms · 2026-07-14T11:48:38.821564+00:00 · methodology

0 comments
read the original abstract

Local motion blur detection requires pixel-level localization of blurred regions. Existing benchmarks let models rely on gradient shortcuts that fail to transfer. We introduce BOCCHI (Blurred Objects Captured across Cameras with Human-annotated Imagery), a real-captured benchmark whose sharp regions overlap the blur gradient distribution and defeat these shortcuts, and propose MSDCT-UNet (Multi-Scale Discrete Cosine Transform UNet), a frequency-aware encoder-decoder injecting multi-scale DCT priors through DCT Attention and FiLM. MSDCT-UNet ranks first in in-domain mIoU and boundary localization on BOCCHI, and BOCCHI-trained models outperform every other training source on cross-dataset transfer with only 633 training images.

Figures

Figures reproduced from arXiv: 2607.10427 by Cheng-Yuan Chiang, Jian-Jiun Ding, Kuan-Lin Chen, Yuan-Kang Lee.

Figure 1
Figure 1. Figure 1: Overview of our contributions. Existing benchmarks can permit gradient-based shortcuts when blur and sharp regions are easily separable. BOCCHI addresses this with 633 real-captured, pixel-annotated images whose sharp regions cover both textured and smooth surfaces, creating strong gradient overlap with blurred objects. Built on this benchmark, MSDCT-UNet injects multi-scale DCT priors and achieves the bes… view at source ↗
Figure 2
Figure 2. Figure 2: Representative samples from BOCCHI (top three rows) and the BOCCHI-protocol subset of the Inference Dataset (bottom row, 164 held-out images for cross-dataset evaluation). Scenes span pedestrians, vehicles, animals, sports, tabletop manipulation, and rail transit [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Two distinguishing properties of BOCCHI across five datasets. (a) Blur-ratio distribution. (b) Per-dataset gradient distribution in blur vs. sharp regions, with sharp-region PR25 and PR50 overlays. Quantitative comparisons and the PR25/µblur ratio are analyzed in Sec. 3 [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Overview of MSDCT-UNet. (Left) DCT feature extraction; (Middle) 4-stage encoder-decoder with NeXtBlock+FiLM fusion and an AFASPP bottle￾neck; (Right) deep supervision over main and three auxiliary heads. 4.2 DCT Feature Extraction Motion blur causes the progressive attenuation of high-frequency components in images. To effectively capture and exploit this intrinsic property, our preprocess￾ing pipeline is … view at source ↗
Figure 5
Figure 5. Figure 5: DCT Attention (single head). Each resolution uses K=4 independent heads with channel-wise dropout (p=0.1) at training only. Together, Eqs. (1) and (2) produce the per-pixel multi-scale DCT spectrum from the gradient map. At each scale we discard the upper-left low-frequency region of CM and keep only the high-frequency (HF) positions, yielding 6, 28, 120, and 496 candidates for M = 3, 7, 15, 31 respectivel… view at source ↗
Figure 6
Figure 6. Figure 6: NeXtBlock + FiLM Fusion. The modulation path (dashed) bypasses the concatenation; Fin and the DCT attention map A are combined as in Eq. (4). Temperature-scaled softmax and weighted sum. For each head, a soft￾max with temperature τ=1.2 is applied along the 57 channel dimension of L g to produce smoother attention weights, which are then used to linearly combine the original DCT channels into a single activ… view at source ↗
Figure 7
Figure 7. Figure 7: Attentive Frequency ASPP. Five spatial branches and a DCT projec￾tion branch concatenated and SE-recalibrated. corresponding encoder stage, and passes the result through another NeXtBlock + FiLM fusion. Each of the four resolutions owns an independent DCTAttention instance (four in total, with non-shared weights): the 57-channel FDCT map is bilinearly resized to that stage’s spatial resolution and fed into… view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative comparison on the BOCCHI validation set. Three representative cases. Predictions are binary masks (black = predicted blur). Full 12-baseline grids and failure cases are in the supplementary. 5.5 Ablation Study [PITH_FULL_IMAGE:figures/full_fig_p013_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Full-baseline qualitative comparison on BOCCHI validation (success cases). Each case occupies two rows: the top row shows Input, GT, and the six older baselines (MSDU-Net, BiSeNetV2, STDC1, STDC2, NAFNet, DDRNet-23); the bottom row shows MSDCT-UNet (ours) and the six newer baselines (Cellpose3, KDSNet-R50, KDSNet-R101, MSDSeg, BEVANet, ESMDL-UNet). Six success cases are stacked vertically. All predictions … view at source ↗
Figure 10
Figure 10. Figure 10: Failure cases on BOCCHI validation (MSDCT-UNet only). Three cases where MSDCT-UNet underperforms; columns are Input, GT, and MSDCT-UNet’s prediction (binary mask) [PITH_FULL_IMAGE:figures/full_fig_p027_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Cross-dataset qualitative comparison. The same MSDCT-UNet architecture trained from scratch on four sources (CUHKmotion, ReLoBlur, OMoBlur, BOCCHI) and evaluated on four Inference Dataset images (no fine￾tuning). Per-cell numbers are IoU; BOCCHI-trained (green box) wins all four cases (0.873, 0.729, 0.920, 0.733). Masks: white = predicted sharp / black = predicted blur [PITH_FULL_IMAGE:figures/full_fig_p… 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

51 extracted references · 19 canonical work pages · 1 internal anchor

  1. [1]

    Scientific Reports15(12 2025).https://doi

    Chen, C., Li, L., Li, B., Li, H., You, Y., Zhou, W., Bin, Y., Wang, Z., Li, J., Zhang, C.: Lightweight unet with multi-module synergy and dual-domain attention for precise skin lesion segmentation. Scientific Reports15(12 2025).https://doi. org/10.1038/s41598-025-28088-1

    work page doi:10.1038/s41598-025-28088-1 2025

  2. [2]

    In: Computer Vi- sion - ECCV 2018

    Chen, L., Zhu, Y., Papandreou, G., Schroff, F., Adam, H.: Encoder-decoder with atrous separable convolution for semantic image segmentation. In: Computer Vi- sion - ECCV 2018. pp. 833–851. Lecture Notes in Computer Science, Springer (2018).https://doi.org/10.1007/978-3-030-01234-2_49

    doi:10.1007/978-3-030-01234-2_49 2018

  3. [3]

    In: Computer Vision - ECCV 2022

    Chen, L., Chu, X., Zhang, X., Sun, J.: Simple baselines for image restoration. In: Computer Vision - ECCV 2022. Lecture Notes in Computer Science, vol. 13667, pp. 17–33. Springer (2022).https://doi.org/10.1007/978-3-031-20071-7_2

    doi:10.1007/978-3-031-20071-7_2 2022

  4. [4]

    In: Vedaldi, A., Bischof, H., Brox, T., Frahm, J

    Cun, X., Pun, C.: Defocus blur detection via depth distillation. In: Vedaldi, A., Bischof, H., Brox, T., Frahm, J. (eds.) Computer Vision - ECCV 2020 - 16th European Conference, Glasgow, UK, August 23-28, 2020, Proceedings, Part XIII. pp. 747–763. Lecture Notes in Computer Science, Springer (2020). https://doi.org/10.1007/978-3-030-58601-0_44,https://doi....

    work page doi:10.1007/978-3-030-58601-0_44 2020

  5. [5]

    In: IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2021

    Fan, M., Lai, S., Huang, J., Wei, X., Chai, Z., Luo, J., Wei, X.: Rethinking bisenet for real-time semantic segmentation. In: IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2021. pp. 9716–9725. Computer Vision Foundation / IEEE (2021).https://doi.org/10.1109/CVPR46437.2021.00959

    doi:10.1109/cvpr46437.2021.00959 2021

  6. [6]

    In: 2017 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2017, Honolulu, HI, USA, July 21-26, 2017

    Golestaneh,S.A.,Karam,L.J.:Spatially-varyingblurdetectionbasedonmultiscale fused and sorted transform coefficients of gradient magnitudes. In: 2017 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2017, Honolulu, HI, USA, July 21-26, 2017. pp. 596–605. IEEE Computer Society (2017).https: //doi.org/10.1109/CVPR.2017.71,https://doi.org/10.11...

    work page doi:10.1109/cvpr.2017.71 2017

  7. [7]

    In: 2017 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2017, Honolulu, HI, USA, July 21-26, 2017

    Gong, D., Yang, J., Liu, L., Zhang, Y., Reid, I.D., Shen, C., van den Hengel, A., Shi, Q.: From motion blur to motion flow: A deep learning solution for removing heterogeneous motion blur. In: 2017 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2017, Honolulu, HI, USA, July 21-26, 2017. pp. 3806–

    2017

  8. [8]

    IEEE Computer Society (2017).https://doi.org/10.1109/CVPR.2017.405, https://doi.org/10.1109/CVPR.2017.405

    work page doi:10.1109/cvpr.2017.405 2017

  9. [9]

    In: Advances in Neural Infor- mation Processing Systems 35, NeurIPS 2022 (2022)

    Guo, M., Lu, C., Hou, Q., Liu, Z., Cheng, M., Hu, S.: Segnext: Rethinking convo- lutional attention design for semantic segmentation. In: Advances in Neural Infor- mation Processing Systems 35, NeurIPS 2022 (2022)

    2022

  10. [10]

    In: 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2018

    Hu, J., Shen, L., Sun, G.: Squeeze-and-excitation networks. In: 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2018. pp. 7132–

    2018

  11. [11]

    IEEE Computer Society (2018).https://doi.org/10.1109/CVPR.2018. 00745

    doi:10.1109/cvpr.2018 2018

  12. [12]

    Hu, Z., Yu, T.: Learning to predict decomposed dynamic filters for single image motiondeblurring.In:AsianConferenceonComputerVision(ACCV).pp.390–408 (2022).https://doi.org/10.1007/978-3-031-26313-2_24

    work page doi:10.1007/978-3-031-26313-2_24 2022

  13. [13]

    In: 2025 IEEE International Conference on Image Processing (ICIP)

    Huang, P.M., Chao, I.T., Huang, P.C., Liao, J.W., Chuang, Y.Y.: Bevanet: Bi- lateral efficient visual attention network for real-time semantic segmentation. In: 2025 IEEE International Conference on Image Processing (ICIP). pp. 2778–2783 (2025).https://doi.org/10.1109/ICIP55913.2025.11084676

    doi:10.1109/icip55913.2025.11084676 2025

  14. [14]

    In: IEEE/CVF International Conference on 16 K.-L

    Huang, Z., Zhang, Z., Lan, C., Zha, Z., Lu, Y., Guo, B.: Adaptive frequency fil- ters as efficient global token mixers. In: IEEE/CVF International Conference on 16 K.-L. Chen et al. Computer Vision, ICCV 2023. pp. 6026–6036. IEEE (2023).https://doi.org/10. 1109/ICCV51070.2023.00556

    arXiv 2023

  15. [15]

    Fractal and Fractional9(7) (2025).https://doi.org/10.3390/ fractalfract9070460,https://www.mdpi.com/2504-3110/9/7/460

    Jeong, S.I., Jeong, M.S., Park, K.R.: Estimation of fractal dimension and se- mantic segmentation of motion-blurred images by knowledge distillation in au- tonomous vehicle. Fractal and Fractional9(7) (2025).https://doi.org/10.3390/ fractalfract9070460,https://www.mdpi.com/2504-3110/9/7/460

    2025

  16. [16]

    In: 2008 IEEE Conference on Computer Vision and Pattern Recognition

    Ji, H., Liu, C.: Motion blur identification from image gradients. In: 2008 IEEE Conference on Computer Vision and Pattern Recognition. pp. 1–8. IEEE (2008)

    2008

  17. [17]

    In: 2021 IEEE/CVF International Conference on Computer Vision, ICCV 2021

    Jiang, L., Dai, B., Wu, W., Loy, C.C.: Focal frequency loss for image reconstruc- tion and synthesis. In: 2021 IEEE/CVF International Conference on Computer Vision, ICCV 2021. pp. 13899–13909. IEEE (2021).https://doi.org/10.1109/ ICCV48922.2021.01366

    arXiv 2021

  18. [18]

    In: 2024 IEEE 13th Global Conference on Consumer Electronics (GCCE)

    Lee, Y.K., Yu, Y.C., Ding, J.J.: Efficient motion blur detection and extraction from a single image in hvs-based gradient domain. In: 2024 IEEE 13th Global Conference on Consumer Electronics (GCCE). pp. 860–861. IEEE (2024)

    2024

  19. [19]

    In: Williams, B., Chen, Y., Neville, J

    Li, H., Zhang, Z., Jiang, T., Luo, P., Feng, H., Xu, Z.: Real-world deep local mo- tion deblurring. In: Williams, B., Chen, Y., Neville, J. (eds.) Thirty-Seventh AAAI Conference on Artificial Intelligence, AAAI 2023, Thirty-Fifth Conference on In- novative Applications of Artificial Intelligence, IAAI 2023, Thirteenth Symposium on Educational Advances in ...

    work page doi:10.1609/aaai.v37i1.25215 2023

  20. [20]

    In: The Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024

    Li, H., Zhao, J., Zhou, S., Feng, H., Li, C., Loy, C.C.: Adaptive window pruning for efficient local motion deblurring. In: The Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024. OpenRe- view.net (2024),https://openreview.net/forum?id=hI18CDyadM

    2024

  21. [21]

    IEEE Trans

    Liu, C., Szeliski, R., Kang, S.B., Zitnick, C.L., Freeman, W.T.: Automatic estima- tion and removal of noise from a single image. IEEE Trans. Pattern Anal. Mach. In- tell.30(2), 299–314 (2008).https://doi.org/10.1109/TPAMI.2007.1176,https: //doi.org/10.1109/TPAMI.2007.1176

    work page doi:10.1109/tpami.2007.1176 2008

  22. [22]

    In: Asian Conference on Computer Vision (ACCV)

    Liu, Y., Yoshie, O., Watanabe, H.: Application of multi-modal fusion attention mechanism in semantic segmentation. In: Asian Conference on Computer Vision (ACCV). pp. 378–397 (2022).https://doi.org/10.1007/978-3-031-26293-7_23

    work page doi:10.1007/978-3-031-26293-7_23 2022

  23. [23]

    In: IEEE/CVF Conference on Computer Vision and Pattern Recog- nition, CVPR 2022

    Liu, Z., Mao, H., Wu, C., Feichtenhofer, C., Darrell, T., Xie, S.: A convnet for the 2020s. In: IEEE/CVF Conference on Computer Vision and Pattern Recog- nition, CVPR 2022. pp. 11966–11976. IEEE (2022).https://doi.org/10.1109/ CVPR52688.2022.01167

    arXiv 2022

  24. [24]

    https://doi.org/10.1007/978-3-030-69532-3_20

    Lumentut, J.S., Santoso, J., Park, I.K.: Human motion deblurring using localized bodyprior.In:AsianConferenceonComputerVision(ACCV).pp.320–335(2020). https://doi.org/10.1007/978-3-030-69532-3_20

    work page doi:10.1007/978-3-030-69532-3_20 2020

  25. [25]

    IEEE Trans

    Ma, K., Fu, H., Liu, T., Wang, Z., Tao, D.: Deep blur mapping: Exploiting high- level semantics by deep neural networks. IEEE Trans. Image Process.27(10), 5155–5166 (2018).https://doi.org/10.1109/TIP.2018.2847421,https://doi. org/10.1109/TIP.2018.2847421

    work page doi:10.1109/tip.2018.2847421 2018

  26. [26]

    In: 2017 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2017, Honolulu, HI, USA, July 21-26, 2017

    Nah, S., Kim, T.H., Lee, K.M.: Deep multi-scale convolutional neural network for dynamic scene deblurring. In: 2017 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2017, Honolulu, HI, USA, July 21-26, 2017. pp. 257–

    2017

  27. [27]

    IEEE Computer Society (2017).https://doi.org/10.1109/CVPR.2017.35, https://doi.org/10.1109/CVPR.2017.35 BOCCHI and MSDCT-UNet for Local Motion Blur Detection 17

    work page doi:10.1109/cvpr.2017.35 2017

  28. [28]

    IEEE Trans

    Pan, H., Hong, Y., Sun, W., Jia, Y.: Deep dual-resolution networks for real-time and accurate semantic segmentation of traffic scenes. IEEE Trans. Intell. Transp. Syst.24(3), 3448–3460 (2023).https://doi.org/10.1109/TITS.2022.3228042

    work page doi:10.1109/tits.2022.3228042 2023

  29. [29]

    In: Proceedings of the Thirty-Second AAAIConferenceonArtificialIntelligence,(AAAI-18).pp.3942–3951.AAAIPress (2018).https://doi.org/10.1609/AAAI.V32I1.11671

    Perez, E., Strub, F., de Vries, H., Dumoulin, V., Courville, A.C.: Film: Visual reasoning with a general conditioning layer. In: Proceedings of the Thirty-Second AAAIConferenceonArtificialIntelligence,(AAAI-18).pp.3942–3951.AAAIPress (2018).https://doi.org/10.1609/AAAI.V32I1.11671

    doi:10.1609/aaai.v32i1.11671 2018

  30. [30]

    In: 30th British Machine Vision Conference 2019, BMVC 2019

    Poudel, R.P.K., Liwicki, S., Cipolla, R.: Fast-scnn: Fast semantic segmentation network. In: 30th British Machine Vision Conference 2019, BMVC 2019. BMVA Press (2019),https://bmvc2019.org/wp-content/uploads/papers/0959-paper. pdf

    2019

  31. [31]

    In: 2021 IEEE/CVF International Conference on Computer Vision, ICCV 2021

    Qin, Z., Zhang, P., Wu, F., Li, X.: Fcanet: Frequency channel attention networks. In: 2021 IEEE/CVF International Conference on Computer Vision, ICCV 2021. pp. 763–772. IEEE (2021).https://doi.org/10.1109/ICCV48922.2021.00082

    doi:10.1109/iccv48922.2021.00082 2021

  32. [32]

    In: Medical Image Computing and Computer-Assisted Intervention - MICCAI 2015

    Ronneberger, O., Fischer, P., Brox, T.: U-net: Convolutional networks for biomed- ical image segmentation. In: Medical Image Computing and Computer-Assisted Intervention - MICCAI 2015. pp. 234–241. Lecture Notes in Computer Science, Springer (2015).https://doi.org/10.1007/978-3-319-24574-4_28

    doi:10.1007/978-3-319-24574-4_28 2015

  33. [33]

    Russell, B.C., Torralba, A., Murphy, K.P., Freeman, W.T.: Labelme: A database andweb-basedtoolforimageannotation.InternationalJournalofComputerVision 77(1), 157–173 (2008).https://doi.org/10.1007/s11263-007-0090-8

    doi:10.1007/s11263-007-0090-8 2008

  34. [34]

    Shi,J.,Xu,L.,Jia,J.:Discriminativeblurdetectionfeatures.In:2014IEEEConfer- ence on Computer Vision and Pattern Recognition, CVPR 2014, Columbus, OH, USA, June 23-28, 2014. pp. 2965–2972. IEEE Computer Society (2014).https: //doi.org/10.1109/CVPR.2014.379,https://doi.org/10.1109/CVPR.2014.379

    doi:10.1109/cvpr.2014.379 2014

  35. [35]

    Nature Methods22(3), 592–599 (2025).https://doi.org/ 10.1038/s41592-025-02595-5,https://doi.org/10.1038/s41592-025-02595-5

    Stringer, C., Pachitariu, M.: Cellpose3: One-click image restoration for improved cellular segmentation. Nature Methods22(3), 592–599 (2025).https://doi.org/ 10.1038/s41592-025-02595-5,https://doi.org/10.1038/s41592-025-02595-5

    doi:10.1038/s41592-025-02595-5 2025

  36. [36]

    In: IEEE Conference on Computer Vision and PatternRecognition,CVPR2015,Boston,MA,USA,June7-12,2015.pp.769–777

    Sun, J., Cao, W., Xu, Z., Ponce, J.: Learning a convolutional neural network for non-uniform motion blur removal. In: IEEE Conference on Computer Vision and PatternRecognition,CVPR2015,Boston,MA,USA,June7-12,2015.pp.769–777. IEEE Computer Society (2015).https://doi.org/10.1109/CVPR.2015.7298677, https://doi.org/10.1109/CVPR.2015.7298677

    doi:10.1109/cvpr.2015.7298677 2015

  37. [37]

    Highly tunable polarization-engineered two-dimensional electron gas in ${\epsilon}$-AlGaO3 / ${\epsilon}$-Ga2O3 heterostructures

    Tong, H., Li, M., Zhang, H., Zhang, C.: Blur detection for digital images using wavelet transform. In: Proceedings of the 2004 IEEE International Conference on Multimedia and Expo, ICME 2004, 27-30 June 2004, Taipei, Taiwan. pp. 17–20. IEEE Computer Society (2004).https://doi.org/10.1109/ICME.2004.1394114, https://doi.org/10.1109/ICME.2004.1394114

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1109/icme.2004.1394114 2004

  38. [38]

    In: Avidan, S., Brostow, G.J., Cissé, M., Farinella, G.M., Hassner, T

    Tsai, F., Peng, Y., Lin, Y., Tsai, C., Lin, C.: Stripformer: Strip transformer for fast image deblurring. In: Avidan, S., Brostow, G.J., Cissé, M., Farinella, G.M., Hassner, T. (eds.) Computer Vision - ECCV 2022 - 17th European Conference, Tel Aviv, Israel, October 23-27, 2022, Proceedings, Part XIX. pp. 146–162. Lec- ture Notes in Computer Science, Sprin...

    work page doi:10.1007/978-3-031-19800-7_9 2022

  39. [39]

    IEEE Trans

    Wang, J., Sun, K., Cheng, T., Jiang, B., Deng, C., Zhao, Y., Liu, D., Mu, Y., Tan, M., Wang, X., Liu, W., Xiao, B.: Deep high-resolution representation learn- ing for visual recognition. IEEE Trans. Pattern Anal. Mach. Intell.43(10), 3349– 3364 (2021).https://doi.org/10.1109/TPAMI.2020.2983686,https://doi.org/ 10.1109/TPAMI.2020.2983686 18 K.-L. Chen et al

    doi:10.1109/tpami.2020.2983686 2021

  40. [40]

    In: Computer Vision - ECCV 2018

    Woo, S., Park, J., Lee, J., Kweon, I.S.: CBAM: convolutional block atten- tion module. In: Computer Vision - ECCV 2018. Lecture Notes in Com- puter Science, vol. 11211, pp. 3–19. Springer (2018).https://doi.org/10.1007/ 978-3-030-01234-2_1

    2018

  41. [41]

    Sensors21(5) (2021).https://doi.org/10.3390/s21051873,https:// www.mdpi.com/1424-8220/21/5/1873

    Xiao, X., Yang, F., Sadovnik, A.: Msdu-net: A multi-scale dilated u-net for blur detection. Sensors21(5) (2021).https://doi.org/10.3390/s21051873,https:// www.mdpi.com/1424-8220/21/5/1873

    work page doi:10.3390/s21051873 2021

  42. [42]

    In: 2020 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, CVPR 2020, Seattle, WA, USA, June 13-19, 2020

    Xu, K., Qin, M., Sun, F., Wang, Y., Chen, Y., Ren, F.: Learning in the fre- quency domain. In: 2020 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, CVPR 2020, Seattle, WA, USA, June 13-19, 2020. pp. 1737–

    2020

  43. [43]

    Computer Vision Foundation / IEEE (2020).https://doi.org/10.1109/ CVPR42600.2020.00181,https://openaccess.thecvf.com/content_CVPR_2020/ html/Xu_Learning_in_the_Frequency_Domain_CVPR_2020_paper.html

    arXiv 2020

  44. [44]

    In: Asian Conference on Computer Vision (ACCV)

    Yang, D., Suh, M.K., Seo, S.W.: AFF-CAM: Adaptive frequency filtering based channel attention module. In: Asian Conference on Computer Vision (ACCV). pp. 373–388 (2022).https://doi.org/10.1007/978-3-031-26351-4_23

    work page doi:10.1007/978-3-031-26351-4_23 2022

  45. [45]

    In: Mathematical Methods in Pattern and Image Analysis

    Yang,K.C.,Guest,C.C.,Das,P.:Motionblurdetectingbysupportvectormachine. In: Mathematical Methods in Pattern and Image Analysis. vol. 5916, pp. 261–273. SPIE (2005)

    2005

  46. [46]

    In: Asian Conference on Computer Vision (ACCV)

    Yang, X., Wu, Y., Zhao, J., Liu, F.: Dense dual-path network for real-time semantic segmentation. In: Asian Conference on Computer Vision (ACCV). pp. 553–570 (2020).https://doi.org/10.1007/978-3-030-69525-5_33

    work page doi:10.1007/978-3-030-69525-5_33 2020

  47. [47]

    Yu, C., Gao, C., Wang, J., Yu, G., Shen, C., Sang, N.: Bisenet V2: bilateral network with guided aggregation for real-time semantic segmentation. Int. J. Comput. Vis. 129(11), 3051–3068 (2021).https://doi.org/10.1007/S11263-021-01515-2

    work page doi:10.1007/s11263-021-01515-2 2021

  48. [48]

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

    Yu, D., Li, J., Zhou, J., Zhuge, Z., Chen, Y., Li, Q.: OMoBlur: An object motion blur dataset and benchmark for real-world local motion deblurring. In: Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 22626–22635 (2026)

    2026

  49. [49]

    In: IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022

    Zamir, S.W., Arora, A., Khan, S., Hayat, M., Khan, F.S., Yang, M.: Restormer: Efficient transformer for high-resolution image restoration. In: IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022. pp. 5718–5729. IEEE (2022).https://doi.org/10.1109/ CVPR52688.2022.00564,https://doi.org/10.1109/...

    doi:10.1109/cvpr52688.2022.00564 2022

  50. [50]

    In: 2017 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2017

    Zhao, H., Shi, J., Qi, X., Wang, X., Jia, J.: Pyramid scene parsing network. In: 2017 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2017. pp. 6230–6239. IEEE Computer Society (2017).https://doi.org/10.1109/CVPR. 2017.660

    doi:10.1109/cvpr 2017

  51. [51]

    Zhao, S., Wang, Y., Huo, Z., Zhang, F.: Lightweight and real-time semantic seg- mentation network via multi-scale dilated convolutions. The Visual Computer41, 11833–11855 (08 2025).https://doi.org/10.1007/s00371-025-04130-1 BOCCHI and MSDCT-UNet for Local Motion Blur Detection 19 Supplementary Material A Implementation Details (Extended) Software environm...

    work page doi:10.1007/s00371-025-04130-1 2025