Pith. sign in

REVIEW 3 major objections 6 minor 68 references

PeCA: Palette Context Assisted Inference for Test-Time Paint-Bucket Colourisation on Animation Videos

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read PeCA claims that paint-bucket colourisation improves at test time by pooling spatial context from expanded reference views, softening top-k correspondences into palette probabilities, and fusing those probabilities only along cycle-consiste

desk verdict A useful, carefully evaluated test-time plug-in for paint-bucket colourisation; the headline DACoN comparison needs an explicit protocol statement before I'd trust the trained-pipeline gains. read the letter →

arxiv 2608.00903 v1 pith:VYOKXEZK submitted 2026-08-02 cs.CV

Dongheng Lin , Jianbo Jiao This is my paper
classification cs.CV
keywords animationcolourisationpaint-bucketregioncorrespondencetest-timeinferencefoundationmodelstemporalconsistencyvideosubmodularmaximization
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

Hand-drawn animation colourisation assigns every line-enclosed region a colour from a fixed character palette, and current pipelines do this by matching each target region to a reference region. The paper argues that failures come less from missing matches than from ambiguous, noisy correspondences, and that context can resolve the ambiguity. It introduces PeCA, a training-free inference wrapper that expands the reference bank with target-aware augmented views, converts noisy top-k matches into per-region palette probabilities via soft voting, and refines those probabilities along bi-directionally consistent matches between adjacent frames. Across design-sheet and first-frame colourisation benchmarks, PeCA reports consistent segment-accuracy gains of roughly 3.9 to 12.1 points on one-shot reference tasks, with the largest gains on frozen foundation backbones that have never seen colourisation training. A sympathetic reader would take this as evidence that context engineering at test time is a viable alternative to training stronger colourisation models.

What carries the argument

The load-bearing object is the per-region palette probability vector: a distribution over the discrete colour palette obtained by soft top-k voting. It makes colour evidence from different views and frames commensurable, which is what lets temporal context act on the same label space. Three mechanisms move evidence into that space: Active Reference Expansion (Eqs. 3-4) scores augmented reference views by the maximum cosine similarity they offer to target regions and selects a budgeted subset via a submodular facility-location objective; Probability Aggregation (Eqs. 5-6) soft-votes top-k reference matches into palette probabilities; Cyclic-gated Temporal Fusion (Eqs. 7-8) keeps only bidirect

What would settle it

Take a character whose left and right sides have different colours, build a target shot in a pose where a horizontally flipped design sheet yields the highest feature similarity, and run PeCA with and without that flip in the candidate pool; if segment accuracy does not improve when the flipped view is selected, the similarity-proxy assumption behind Active Reference Expansion fails. A direct quantitative check is the GT-Margin of Eq. (10): if adding ARE-selected views does not raise the probability gap between the ground-truth colour and its strongest competitor, the expansion is supplying di

Watch

Extended reading notes

Core claim

The paper's central claim is that region-matching colourisation becomes more robust without task-specific training when evidence is pooled from three kinds of context. Active Reference Expansion augments design sheets with geometric transforms and greedily selects a budgeted subset that best covers target regions via a submodular facility-location objective. Probability Aggregation replaces hard top-1 copying with temperature-sharpened soft top-k voting into the shared colour-palette simplex, giving per-region colour probabilities comparable across frames. Cyclic-gated Temporal Fusion refines those probabilities only along bidirectional cycle-consistent matches between adjacent frames. The p

Load-bearing premise

The load-bearing premise is that the maximum cosine similarity between a target region's average-pooled descriptor and an augmented reference view's descriptors is a reliable proxy for the colour support that view provides; if geometric augmentation creates high-similarity distractors with wrong region identity, the selected views could add noise rather than evidence.

Editorial extensions

If this is right

  • Frozen foundation models become usable paint-bucket colourisers out of the box: SAM2.1-Large's one-shot design-sheet segment accuracy rises from 34.54% to 46.65% when PeCA is wrapped around it.
  • The gains do not saturate when more design sheets are available: 5-shot and max-shot reference protocols still show improvements, contrasting with the earlier observation that extra references give marginal returns.
  • Longer videos are handled without accumulating error: on the new 206-frame Anita-Pirate benchmark, PeCA improves accuracy and raises a temporal-consistency F1 from about 91% to 97%.
  • The same inference wrapper transfers to non-colour label propagation: on VIPSeg superpixel semantic labelling, PeCA improves segment accuracy and mIoU over direct matching with two frozen backbones.
  • Because PeCA is plug-and-play, it can be combined with future region-matching backbones without retraining, so descriptor-quality improvements and test-time context improvements are additive.

Reading between the lines

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

  • The dependence on geometric augmentation means PeCA reduces mild in-plane pose and layout gaps but not out-of-plane rotations or never-observed character poses; a natural extension is to pair the same facility-location selection with learned novel-view generation instead of handcrafted transforms.
  • The active reference selection objective could be inverted into an interactive production tool: instead of expanding views, the artist could be told which few frames or design-sheet views to colour first, minimising manual effort while maximising downstream automatic colourisation accuracy.
  • The palette-probability interface is task-agnostic; any discrete-label assignment problem with per-region descriptors and a fixed label set could adopt the same soft-voting plus cycle-gated temporal fusion, with the VIPSeg result as a first indication.
  • A stricter test of the temporal cue would be a video in which colours legitimately change over time, such as a lighting shift or palette swap; PeCA's cycle gate should refuse fusion there, and measuring that refusal would bound the method's applicability.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper introduces PeCA, a training-free, plug-and-play inference framework for paint-bucket colourisation in animation videos. PeCA builds on segment-level retrieval: region descriptors are obtained by pooling backbone features, and correspondences are computed by cosine similarity. The framework adds three test-time components: Active Reference Expansion (ARE), which augments reference design sheets and selects a target-conditioned support subset via a submodular facility-location objective; Probability Aggregation (PA), which soft-votes over top-k correspondences to obtain per-region palette probabilities; and Cyclic-gated Temporal Fusion (CT), which refines these probabilities along cycle-consistent links between adjacent frames. Experiments on PBC-3D, PBC-Real, and a newly constructed Anita-Pirate long-video test set report consistent gains over a top-1 retrieval baseline on the trained DACoN 1.1 pipeline and on several frozen foundation backbones (DINOv2, SAM2.1, SigLIPv2, DINOv3, CLIP).

Significance. If the reported gains are faithful, PeCA is a useful contribution: it is model-agnostic, requires no task-specific training, and the paper provides unusually thorough validation for a test-time inference method, including ablations of each component (Tables 5-6), hyperparameter sensitivity surfaces (Fig. 7), multiple backbones and reference protocols, and a new long-video stress test with honest failure-case analysis in App. H. The circularity concern raised in the stress-test note is only partially supported: ARE and PA do use the same similarity scores, but Table 5 shows ARE alone contributes about one point, so the method does not hinge on that assumption. The main unresolved issues are the verification of the DACoN 1.1 baseline and the absence of statistical uncertainty; these are load-bearing for the headline claim that PeCA improves a trained SOTA pipeline.

major comments (3)
  1. [Sec. 4.1, Eq. (2), Tables 1-4] The claim that PeCA improves the trained DACoN 1.1 pipeline is not yet verifiable from the manuscript. The implementation details state that the authors 'build a Base inference for SOTA [29] and frozen foundation models', and the base rule is defined as the top-1 retrieval in Eq. (2). It is never stated explicitly that the rows labelled 'DACoN 1.1' are produced by the official DACoN 1.1 code/checkpoint and inference protocol, rather than by a re-implemented top-1 retrieval baseline using DACoN features. If the latter is the case, then PeCA's probability aggregation (Eq. 6) is replacing the only inference rule of that baseline, and a substantial fraction of the reported gain (e.g., 68.01 to 72.04 in Table 1) could be due to recovering the gap between simple top-1 retrieval and the actual DACoN inference. No code or checkpoints are released, so this cannot be checked. Please clarify exactl
  2. [Tables 1-6, App. B.3] All reported metrics are single-run percentages with no variance, confidence intervals, or significance tests. This matters because ARE is stochastic: App. B.3 and Table B2 sample flips, rotations, and affine transformations. A single draw cannot support claims of 'consistent improvements', especially for small gains such as Pix-B-MIoU improvements of 0.31-0.82 points (Tables 1-2). Please report mean and standard deviation over multiple runs or seeds, and ideally a paired significance test for the headline comparisons.
  3. [Sec. 4.1, Fig. 7] The fixed hyperparameters (top-k=64, tau=0.05, B=31, m=4) are said to be used across all experiments, but the manuscript does not describe any validation split or selection criterion used to choose these values. Without this information, it is unclear whether the defaults were selected on the test benchmarks, which would inflate the reported gains. Please specify how the hyperparameters were chosen, or evaluate sensitivity on a held-out validation set rather than on the final test sets.
minor comments (6)
  1. [Sec. 3.4, Eq. (5)] The sentence 'When k=1 and tau->0, it reduces to hard copying' is imprecise: for k=1 the softmax is degenerate for any tau. Consider reformulating as 'when only the top-1 candidate is retained'.
  2. [App. B.3, Eq. (vi) and Table B2] The transform composition is unclear: is the chain T_affine * T_90 * T_vflip * T_hflip applied with each probability independently, or is a single composition sampled? Please clarify the sampling procedure.
  3. [App. F] The VIPSeg label-propagation experiment selects reference frames using ground-truth panoptic classes present in the target video. This is an oracle protocol; please state explicitly that these results are an upper-bound diagnostic rather than a deployable setting.
  4. [Table 4] On Anita-Pirate, DACoN 1.1 + PeCA has the same Pix-Acc as DACoN 1.1 (94.29). The text says the framework provides 'consistent improvements'; please acknowledge tied metrics when they occur.
  5. [App. A] The Anita-Pirate benchmark is constructed semi-automatically by inferring shadow/highlight boundaries from colour discontinuities and assigning majority colours. Please include a discussion of potential label noise and how it may affect conclusions when this benchmark is used as a stress test.
  6. [Overall] No code or official checkpoints are released, so exact reproduction is difficult; the project page appears to host only the new test set. A statement on planned code release would strengthen reproducibility.

Circularity Check

1 steps flagged · score 2.0 of 10

No central circularity; one appendix temporal-stability metric is defined from the method's own cycle-consistency gate, giving a partially self-confirming diagnostic.

  1. self definitional [Appendix E, Eqs. (xi)-(xv); cf. Sec. 3.5 Eqs. (7)-(8)]
    "we evaluate temporal stability on surrogate temporal correspondences, following the same strategy used in CT. ... we further retain only cycle-consistent links as stable: stable(t,i)=1[i=arg max_{i'} <f_{t-1,j*_t(i)}, f_{t,i'}>]. (xii) ... TM-F1_t = 2 P_t R_t / (P_t + R_t). (xv)"

    The metric's stable-link filter (Eq. xii) is the same cycle-consistency test as Eq. (7) (rho_t(pi_t(i))=i), and the CT module being evaluated applies its product fusion (Eq. 8) only along exactly those links. Thus the predictions are shaped to maintain label agreement on precisely the links that TM-F1 scores, so the reported temporal-stability improvement is partly produced by the metric's own definition rather than by an independent measure. This is confined to the appendix diagnostic and does not make the main accuracy tables circular.

full rationale

PeCA is an inference-time procedure with no trainable parameters, so the central colorisation claim (Tables 1-4) is not a fitted-input-called-prediction reduction: the per-region palette probabilities (Eqs. 5-6) are deterministic functions of reference palette labels, and the reported segment/pixel accuracies are evaluated against external ground truth. The active reference expansion (Eqs. 3-4) uses the same region-descriptor similarities that later drive matching, but it optimizes a coverage objective over reference views, not the colour labels, so it is a transductive selection heuristic rather than a circular definition. No load-bearing self-citations or imported uniqueness theorems appear. The one genuine circular element is the Appendix E temporal-stability metric, which filters evaluated links by the same cycle-consistency condition that the temporal fusion module uses to decide where to propagate colours; this makes the TM-F1 diagnostic partially self-confirming. The unverified 'DACoN 1.1 Base' comparison is a reproducibility concern, not a circularity. Overall score 2.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

PeCA introduces no new physical or model entities. Its free parameters are standard inference hyperparameters (top-k, temperature, view budget, exploration factor, subsample size) and the augmentation distribution. The main domain assumptions are the flood-fill closed-region assumption and the reliability of backbone feature similarities as region identity proxies, both shared with prior segment-matching colourisation work.

free parameters (6)
  • top-k = 64
    Number of candidate reference regions used in soft voting (Eq. 5-6). Chosen by sensitivity analysis (Fig. 7a-b); affects selectivity vs. dilution.
  • temperature tau = 0.05
    Sharpness of soft-vote weights in Eq. (5). Chosen together with top-k; smaller tau approaches hard top-1, larger tau approaches global mixing.
  • num selected views B = 31
    Budget for active reference expansion in Eq. (4). Selected as a cost-effective operating point in Fig. 7c-d.
  • exploration factor m = 4
    Candidate pool size is mB for reference expansion; m=4 means 124 candidates generated and 31 selected. Chosen from Fig. 7c-d.
  • target frame subsample size Ts = 20
    Number of uniformly sampled target frames used to evaluate the facility-location objective in Eq. (4). Bounds selection cost (App. B.3).
  • ARE transformation distribution = hflip p=0.5, vflip p=0.1, rot p=0.2, affine angle U[-30,30], scale U[0.5,2.0], translation up to 50%
    Parameters of the geometric augmentation distribution in Tab. B2. Fixed across all experiments; not justified by a separate selection procedure.
assumptions (4)
  • domain assumption Flood-fill segmentation on production line sketches yields closed, meaningful regions corresponding to paint-bucket areas.
    Inherited from the paint-bucket formulation; App. H explicitly admits failure on amateur sketches with line leakage where flood-fill may leak or over-segment.
  • domain assumption Geometric augmentations applied jointly to the line, segment map, and colour image preserve region identity and palette labels.
    Sec. B.3 applies flips, rotations, and affine warps with scale up to 2x to the reference triplet; the colour label of each transformed region is assumed valid for matching. Large distortions could break this.
  • domain assumption Cosine similarity between region-pooled frozen backbone features is a valid proxy for region identity across appearance gaps.
    Central to Eqs. (1)-(4) for reference matching and Eq. (7) for temporal cycle checks. The method inherits the quality of the backbone descriptors.
  • standard math The facility-location objective F(V) is monotone submodular, so greedy selection provides a near-optimal approximation.
    Invoked in Sec. 3.3 with citation [17]; this is a standard result for monotone submodular maximization with cardinality constraints.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PeCA: Palette Context Assisted Inference for Test-Time Paint-Bucket Colourisation on Animation Videos." pith.science (2026). https://pith.science/paper/VYOKXEZK

@misc{pith2026260800903,
  author       = {Pith},
  title        = {Pith review of: PeCA: Palette Context Assisted Inference for Test-Time Paint-Bucket Colourisation on Animation Videos},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VYOKXEZK}},
  note         = {Machine review of arXiv:2608.00903}
}
read the original abstract

In animation production, paint-bucket colourisation for hand-drawn animation is a labour-intensive procedure that assigns each enclosed region in line sketches a colour from reference design sheets. Recent automatic paint-bucket colourisation pipelines mirror this workflow via region correspondence, but correspondences can be brittle when regions are ambiguous fragments without proper context. In this paper, we propose Palette Context Assisted (PeCA), a new training-free, plug-and-play framework for animation video colourisation that aims to close this gap at test-time via reasoning over spatial and temporal contexts. Extensive experiments on existing benchmarks and a newly introduced long-video test case show consistent performance boosts.

Figures

Figures reproduced from arXiv: 2608.00903 by the authors.

Figure 1
Figure 1. Conceptual overview of PeCA. Compared to a baseline that colours re￾gions using isolated correspondences, PeCA leverages temporal and spatial context to pick supports for a more reliable colourisation. Existing automated colourisation approaches can be broadly split into pixel￾based generative models and segment-based pipelines. Pixel-generative meth￾ods [5, 13, 22, 40, 44, 56, 63, 67], including recent DiT-based ge… view at source ↗
Figure 2
Figure 2. The proposed PeCA framework overview. Active Reference Expan￾sion (Sec. 3.3) builds a target-conditioned reference support set. Probability Aggre￾gation (Sec. 3.4) aggregates noisy matches into per-region palette colour probabilities via soft voting. Cyclic-gated Temporal Fusion (Sec. 3.5) fuses colour probabilities across adjacent frames through cycle-consistent temporal links, altogether improving colourisation wi… view at source ↗
Figure 3
Figure 3. Active Reference Expan￾sion. Regions are encoded as features (◦). There are target segments that hardly match the reference (◦). Se￾lecting Target-conditioned views (▶), can maximise coverage to targets, while unselected views ◀ provide lim￾ited support to target video. However, naive test-time augmentation is unlikely to scale efficiently at inference time under this setting [46]. In basic infer￾ence pipelines [7, … view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Cyclic-Gated Temporal Fu￾sion utilises temporal context by fus￾ing per-region colour probabilities only along cycle-consistent matches between adjacent frames, refining colours while avoiding unreliable temporal fusions. Now each frame has per-region palette￾colour pro…
Figure 5
Figure 5. Figure 5: Qualitative comparison under two production formulations. [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: Qualitative results of in-between colourisation on Anita-Pirate. [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]
Figure 7
Figure 7. Figure 7: Hyperparameter sensitivity for PeCA inference under key-frame (one-shot design-sheet referenced) colourisation. We visualise the sensitivity of (top-k, τ ) in PA (Sec. 3.4) and (#views B vs. m) in ARE Sec. 3.3, reported under Segment Accuracy and Pixel Accuracy. Ablati…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

68 extracted references · 47 canonical work pages

  1. [29]

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

    Nagata, K., Kaneko, N.: DACoN: Dino for anime paint bucket colorization with any number of reference images. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 17899–17908 (2025)

    work page 2025

  2. [1]

    https://zhenglinpan.github.io/AnitaDataset_homepage/, accessed: 2024-06-24

    Anita dataset. https://zhenglinpan.github.io/AnitaDataset_homepage/, accessed: 2024-06-24

    work page 2024

  3. [2]

    Technical report, EPFL (06 2010)

    Achanta, R., Shaji, A., Smith, K., Lucchi, A., Fua, P., Süsstrunk, S.: SLIC super- pixels. Technical report, EPFL (06 2010)

    work page 2010

  4. [3]

    In: SIGGRAPH Asia 2024 Technical Communications

    Allen, B., Maejima, A., Anjyo, K.: Fast leak-resistant segmentation for anime line art. In: SIGGRAPH Asia 2024 Technical Communications. SA ’24, Association for Computing Machinery, New York, NY, USA (2024).https://doi.org/10.1145/ 3681758.3698003,https://doi.org/10.1145/3681758.3698003

    arXiv 2024

  5. [4]

    In: Computer Graphics Forum

    Cao, R., Mo, H., Gao, C.: Line art colorization based on explicit region segmen- tation. In: Computer Graphics Forum. vol. 40, pp. 1–10. Wiley Online Library (2021)

    work page 2021

  6. [5]

    IEEE Transactions on Visualization and Computer Graphics 30(10), 6956–6969 (2024)

    Cao, Y., Meng, X., Mok, P., Lee, T.Y., Liu, X., Li, P.: AnimeDiffusion: Anime diffusion colorization. IEEE Transactions on Visualization and Computer Graphics 30(10), 6956–6969 (2024)

    work page 2024

  7. [6]

    In: proceedings of the IEEE Conference on Computer Vision and Pattern Recognition

    Carreira, J., Zisserman, A.: Quo vadis, action recognition? a new model and the kinetics dataset. In: proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. pp. 6299–6308 (2017) PeCA: Palette Context Assisted Inference 33

    work page 2017

  8. [7]

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

    Casey, E., Pérez, V., Li, Z.: The animation transformer: Visual correspondence via segment matching. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 11323–11332 (2021)

    work page 2021

Show all 68 references
  1. [8]

    arXiv preprint arXiv:2410.19424 (2024)

    Dai, Y., Li, Q., Zhou, S., Luo, Y., Li, C., Loy, C.C.: Paint bucket colorization using anime character color design sheets. arXiv preprint arXiv:2410.19424 (2024)

    2024 arXiv

  2. [9]

    In: Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition

    Dai, Y., Zhou, S., Li, Q., Li, C., Loy, C.C.: Learning inclusion matching for ani- mation paint bucket colorization. In: Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition. pp. 25544–25553 (2024)

    2024

  3. [10]

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

    Feng, X., Huang, T., Wang, P., Huang, Z., Haihang, Z., Zou, Y., Li, D., Zou, K.: A unified framework for industrial cel-animation colorization with temporal- structural awareness. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 19301–19310 (Oct...

    2025

  4. [11]

    Google: Nano Banana 2: Combining Pro capabilities with lightning-fast speed — blog.google.https://blog.google/innovation- and- ai/technology/ai/nano- banana-2/, [Accessed 21-06-2026]

    2026

  5. [12]

    In: ACM SIGGRAPH 2024 Posters, pp

    Guajardo, J., Bursalioglu, O., Goldman, D.B.: Generative ai for 2d character ani- mation. In: ACM SIGGRAPH 2024 Posters, pp. 1–2 (2024)

    2024

  6. [13]

    ACM Transactions on Graphics (TOG)43(6), 1–11 (2024)

    Huang, Z., Zhang, M., Liao, J.: LVCD: reference-based lineart video colorization with diffusion models. ACM Transactions on Graphics (TOG)43(6), 1–11 (2024)

    2024

  7. [14]

    In: European Conference on Computer Vision

    Karaev, N., Rocco, I., Graham, B., Neverova, N., Vedaldi, A., Rupprecht, C.: CoTracker: It is better to track together. In: European Conference on Computer Vision. pp. 18–35. Springer (2024)

    2024

  8. [15]

    In: Proceedings of the AAAI Conference on Artificial Intel- ligence

    Kim, S., Park, D., Shim, B.: Semantic-aware superpixel for weakly supervised se- mantic segmentation. In: Proceedings of the AAAI Conference on Artificial Intel- ligence. vol. 37, pp. 1142–1150 (2023)

    2023

  9. [16]

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

    Kirillov, A., Mintun, E., Ravi, N., Mao, H., Rolland, C., Gustafson, L., Xiao, T., Whitehead, S., Berg, A.C., Lo, W.Y., et al.: Segment anything. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 4015–4026 (2023)

    2023

  10. [17]

    Tractability3(71- 104), 3 (2014)

    Krause, A., Golovin, D.: Submodular function maximization. Tractability3(71- 104), 3 (2014)

    2014

  11. [18]

    Li, J., Liang, Q., Li, Q., Gang, R., Fang, J., Lin, C., Feng, S., Liu, X.: RTTLC: Video colorization with restored transformer and test-time local converter. pp. 1722–1730 (06 2023).https://doi.org/10.1109/CVPRW59228.2023.00173

    2023

  12. [19]

    arXiv preprint arXiv:2508.10881 (2025)

    Li, L., Wang, G., Zhang, Z., Li, Y., Li, X., Dou, Q., Gu, J., Xue, T., Shan, Y.: Tooncomposer: Streamlining cartoon production with generative post-keyframing. arXiv preprint arXiv:2508.10881 (2025)

    2025 arXiv

  13. [20]

    In: Proceedings of the International AAAI Conference on Web and Social Media

    Liu, R., Pei, J., Zhu, J.: From pixels to personas: Tracking the evolution of anime characters. In: Proceedings of the International AAAI Conference on Web and Social Media. vol. 20, pp. 1488–1504 (2026)

    2026

  14. [21]

    In: 2023 IEEE/CVF International Con- ference on Computer Vision

    Liu, R., Wu, R., Van Hoorick, B., Tokmakov, P., Zakharov, S., Vondrick, C.: Zero- 1-to-3: Zero-shot one image to 3d object. In: 2023 IEEE/CVF International Con- ference on Computer Vision. pp. 9264–9275. IEEE (2023)

    2023

  15. [22]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Liu, Z., Cheng, K.L., Chen, X., Xiao, J., Ouyang, H., Zhu, K., Liu, Y., Shen, Y., Chen, Q., Luo, P.: Manganinja: Line art colorization with precise reference follow- ing. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 5666–5677 (2025)

    2025

  16. [23]

    Lowe, D.G.: Distinctive image features from scale-invariant keypoints. Int. J. Comput. Vision60(2), 91–110 (Nov 2004).https://doi.org/10.1023/B:VISI. 0000029664.99615.94,https://doi.org/10.1023/B:VISI.0000029664.99615. 94 34 D. Lin and J. Jiao

    2004

  17. [24]

    In: ACM SIGGRAPH 2019 Posters, pp

    Maejima, A., Kubo, H., Funatomi, T., Yotsukura, T., Nakamura, S., Mukaigawa, Y.: Graph matching based anime colorization with multiple references. In: ACM SIGGRAPH 2019 Posters, pp. 1–2 (2019)

    2019

  18. [25]

    In: SIG- GRAPH Asia 2021 Technical Communications, pp

    Maejima, A., Kubo, H., Shinagawa, S., Funatomi, T., Yotsukura, T., Nakamura, S., Mukaigawa, Y.: Anime character colorization using few-shot learning. In: SIG- GRAPH Asia 2021 Technical Communications, pp. 1–4 (2021)

    2021

  19. [26]

    In: Advances in Neural Information Processing Systems (2022)

    Manli, S., Weili, N., De-An, H., Zhiding, Y., Tom, G., Anima, A., Chaowei, X.: Test-time prompt tuning for zero-shot generalization in vision-language models. In: Advances in Neural Information Processing Systems (2022)

    2022

  20. [27]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Meng, Y., Ouyang, H., Wang, H., Wang, Q., Wang, W., Cheng, K.L., Liu, Z., Shen, Y., Qu, H.: AniDoc: Animation creation made easier. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 18187–18197 (2025)

    2025

  21. [28]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (2022)

    Miao, J., Wang, X., Wu, Y., Li, W., Zhang, X., Wei, Y., Yang, Y.: Large-scale video panoptic segmentation in the wild: A benchmark. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (2022)

    2022

  22. [30]

    Nakanishi, H., Shichijo, N., Sugi, M., Ogata, T., Hara, T., Ota, J.: Modeling the process of animation production. Int. J. Autom. Technol.7(4), 439–450 (2013)

    2013

  23. [31]

    arXiv preprint arXiv:2304.07193 (2023)

    Oquab, M., Darcet, T., Moutakanni, T., Vo, H., Szafraniec, M., Khalidov, V., Fernandez,P.,Haziza,D.,Massa,F.,El-Nouby,A.,etal.:DINOv2:Learningrobust visual features without supervision. arXiv preprint arXiv:2304.07193 (2023)

    2023 arXiv

  24. [32]

    arXiv preprint arXiv:2212.09748 (2022)

    Peebles, W., Xie, S.: Scalable diffusion models with transformers. arXiv preprint arXiv:2212.09748 (2022)

    2022 arXiv

  25. [33]

    ACM Transactions on Graphics (TOG)43(4), 1–13 (2024)

    Peng, H.Y., Zhang, J.P., Guo, M.H., Cao, Y.P., Hu, S.M.: Charactergen: Efficient 3d character generation from single images with multi-view pose canonicalization. ACM Transactions on Graphics (TOG)43(4), 1–13 (2024)

    2024

  26. [34]

    arXiv preprint arXiv:2511.04460 (2025)

    Qiao, R., Tan, Q., Yang, M., Dong, G., Yang, P., Lang, S., Wan, E., Wang, X., Xu, Y., Yang, L., et al.: V-thinker: Interactive thinking with images. arXiv preprint arXiv:2511.04460 (2025)

    2025

  27. [35]

    In: International Conference on Machine Learning

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al.: Learning transferable visual models from natural language supervision. In: International Conference on Machine Learning. pp. 8748–8763. PMLR (2021)

    2021

  28. [36]

    Journal of machine learning research11(sept), 2487–2531 (2010)

    Radovanovic, M., Nanopoulos, A., Ivanovic, M.: Hubs in space: Popular nearest neighbors in high-dimensional data. Journal of machine learning research11(sept), 2487–2531 (2010)

    2010

  29. [37]

    In: SIGGRAPH Asia 2018 Posters, pp

    Ramassamy, S., Kubo, H., Funatomi, T., Ishii, D., Maejima, A., Nakamura, S., Mukaigawa, Y.: Pre-and post-processes for automatic colorization using a fully convolutional network. In: SIGGRAPH Asia 2018 Posters, pp. 1–2 (2018)

    2018

  30. [38]

    arXiv preprint arXiv:2408.00714 (2024)

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

    2024 arXiv

  31. [39]

    In: Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition

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

    2022

  32. [40]

    arXiv preprint arXiv:2601.00296 (2026) PeCA: Palette Context Assisted Inference 35

    Sadihin, B.C., Meng, Y., Wang, M.H., Chen, M.J., Su, H.: TimeColor: Flexible reference colorization via temporal concatenation. arXiv preprint arXiv:2601.00296 (2026) PeCA: Palette Context Assisted Inference 35

    2026 arXiv

  33. [41]

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

    Sasaki, K., Iizuka, S., Simo-Serra, E., Ishikawa, H.: Joint gap detection and in- painting of line drawings. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 5725–5733 (2017)

    2017

  34. [42]

    arXiv preprint arXiv:1806.02705 (2018)

    Schuurmans, M., Berman, M., Blaschko, M.B.: Efficient semantic image segmen- tation with superpixel pooling. arXiv preprint arXiv:1806.02705 (2018)

    2018 arXiv

  35. [43]

    In: Proceedings of the IEEE/CVF International Confer- ence on Computer Vision

    Shanmugam, D., Blalock, D., Balakrishnan, G., Guttag, J.: Better aggregation in test-time augmentation. In: Proceedings of the IEEE/CVF International Confer- ence on Computer Vision. pp. 1214–1223 (2021)

    2021

  36. [44]

    IEEE Transactions on Visualization and Computer Graphics29(6), 2965–2979 (2022)

    Shi, M., Zhang, J.Q., Chen, S.Y., Gao, L., Lai, Y.K., Zhang, F.L.: Reference-based deep line art video colorization. IEEE Transactions on Visualization and Computer Graphics29(6), 2965–2979 (2022)

    2022

  37. [45]

    In: Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition

    Shlapentokh-Rothman, M., Blume, A., Xiao, Y., Wu, Y., TV, S., Tao, H., Lee, J.Y., Torres, W., Wang, Y.X., Hoiem, D.: Region-based representations revisited. In: Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition. pp. 17107–17116 (2024)

    2024

  38. [46]

    Journal of big data6(1), 1–48 (2019)

    Shorten, C., Khoshgoftaar, T.M.: A survey on image data augmentation for deep learning. Journal of big data6(1), 1–48 (2019)

    2019

  39. [47]

    Siméoni, O., Vo, H.V., Seitzer, M., Baldassarre, F., Oquab, M., Jose, C., Khalidov, V., Szafraniec, M., Yi, S., Ramamonjisoa, M., Massa, F., Haziza, D., Wehrstedt, L., Wang, J., Darcet, T., Moutakanni, T., Sentana, L., Roberts, C., Vedaldi, A., Tolan, J., Brandt, J., Couprie, ...

    2025 arXiv

  40. [48]

    In: Proceedings of the 6th Annual Conference on Computer Graphics and Interactive Techniques

    Smith, A.R.: Tint fill. In: Proceedings of the 6th Annual Conference on Computer Graphics and Interactive Techniques. pp. 276–283. SIGGRAPH ’79, Association for Computing Machinery, New York, NY, USA (1979).https://doi.org/10. 1145/800249.807456,https://doi.org/10.1145/800249.807456

    1979

  41. [49]

    Advances in Neural Information Processing Systems 36, 1363–1389 (2023)

    Tang, L., Jia, M., Wang, Q., Phoo, C.P., Hariharan, B.: Emergent correspon- dence from image diffusion. Advances in Neural Information Processing Systems 36, 1363–1389 (2023)

    2023

  42. [50]

    arXiv preprint arXiv:2501.06250 (2025)

    Tang, Y., Guo, J., Liu, P., Wang, Z., Hua, H., Zhong, J.X., Xiao, Y., Huang, C., Song, L., Liang, S., et al.: Generative AI for cel-animation: A survey. arXiv preprint arXiv:2501.06250 (2025)

    2025

  43. [51]

    In: European conference on computer vision

    Teed, Z., Deng, J.: Raft: Recurrent all-pairs field transforms for optical flow. In: European conference on computer vision. pp. 402–419. Springer (2020)

    2020

  44. [52]

    arXiv preprint arXiv:2502.14786 (2025)

    Tschannen, M., Gritsenko, A., Wang, X., Naeem, M.F., Alabdulmohsin, I., Parthasarathy, N., Evans, T., Beyer, L., Xia, Y., Mustafa, B., et al.: SIGLIP 2: Multilingual vision-language encoders with improved semantic understanding, lo- calization, and dense features. arXiv prepri...

    2025 arXiv

  45. [53]

    Wang, D., Shelhamer, E., Liu, S., Olshausen, B., Darrell, T.: Tent: Fully test- timeadaptationbyentropyminimization.In:InternationalConferenceonLearning Representations (2021),https://openreview.net/forum?id=uXl3bZLkr3c

    2021

  46. [54]

    In: Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition

    Wang, X., Jabri, A., Efros, A.A.: Learning correspondence from the cycle- consistency of time. In: Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition. pp. 2566–2576 (2019)

    2019

  47. [55]

    University of California Press (1977),https: //books.google.co.uk/books?id=xQhbwgEACAAJ

    Wittgenstein, L.: Remarks on Colour. University of California Press (1977),https: //books.google.co.uk/books?id=xQhbwgEACAAJ

    1977

  48. [56]

    ACM Transactions on Graphics (TOG)43(6), 1–11 (2024) 36 D

    Xing, J., Liu, H., Xia, M., Zhang, Y., Wang, X., Shan, Y., Wong, T.T.: Tooncrafter: Generative cartoon interpolation. ACM Transactions on Graphics (TOG)43(6), 1–11 (2024) 36 D. Lin and J. Jiao

    2024

  49. [57]

    IEEE transactions on pattern analysis and machine intelligence45(1), 285–312 (2022)

    Xu, P., Hospedales, T.M., Yin, Q., Song, Y.Z., Xiang, T., Wang, L.: Deep learning for free-hand sketch: A survey. IEEE transactions on pattern analysis and machine intelligence45(1), 285–312 (2022)

    2022

  50. [58]

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

    Yang, Y., Fan, L., Lin, Z., Wang, F., Zhang, Z.: Layeranimate: Layer-level con- trol for animation. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). pp. 10865–10874 (October 2025)

    2025

  51. [59]

    In: Euro- pean Conference on Computer Vision (ECCV) (2020)

    Zhang, L., JI, Y., Liu, C.: Danbooregion: An illustration region dataset. In: Euro- pean Conference on Computer Vision (ECCV) (2020)

    2020

  52. [60]

    arXiv preprint arXiv:2111.03930 (2021)

    Zhang, R., Fang, R., Zhang, W., Gao, P., Li, K., Dai, J., Qiao, Y., Li, H.: Tip-adapter: Training-free clip-adapter for better vision-language modeling. arXiv preprint arXiv:2111.03930 (2021)

    2021 arXiv

  53. [61]

    Zhang, X., Du, H., Wei, X., Li, Q.: Omnicolor: A unified framework for multi-modal lineart colorization (2026),https://arxiv.org/abs/2603.27531

    2026

  54. [62]

    Zhang, Y., Ma, Y., Wang, B., Chen, Q., Wang, Z.: Follow-your-color: Multi- instance sketch colorization (2025),https://arxiv.org/abs/2503.16948

    2025 arXiv

  55. [63]

    In: Proceed- ings of the 33rd ACM International Conference on Multimedia

    Zhang, Y., Wang, L., Wang, H., Wu, D., Lin, Z., Wang, F., Song, L.: Animecolor: Reference-based animation colorization with diffusion transformers. In: Proceed- ings of the 33rd ACM International Conference on Multimedia. pp. 6682–6690 (2025)

    2025

  56. [64]

    In: The Twelfth International Conference on Learning Repre- sentations

    Zhao, H., Cai, Z., Si, S., Ma, X., An, K., Chen, L., Liu, Z., Wang, S., Han, W., Chang, B.: MMICL: Empowering vision-language model with multi-modal in-context learning. In: The Twelfth International Conference on Learning Repre- sentations

  57. [65]

    In: 2023 IEEE International Conference on Image Processing (ICIP)

    Zhao, Y., Zheng, H., Luo, J., Lam, E.Y.: Improving video colorization by test-time tuning. In: 2023 IEEE International Conference on Image Processing (ICIP). pp. 166–170. IEEE (2023)

    2023

  58. [66]

    arXiv preprint arXiv:2405.07425 (2024)

    Zhenglin Pan, Yu Zhu, Y.M.: Sakuga-42m dataset: Scaling up cartoon research. arXiv preprint arXiv:2405.07425 (2024)

    2024 arXiv

  59. [67]

    arXiv preprint arXiv:2412.11815 (2024)

    Zhuang, J., Ju, X., Zhang, Z., Liu, Y., Zhang, S., Yuan, C., Shan, Y.: ColorFlow: Retrieval-augmented image sequence colorization. arXiv preprint arXiv:2412.11815 (2024)

    2024 arXiv

  60. [68]

    In: Proceedings of the Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers

    Zhuang, J., Li, L., Ju, X., Zhang, Z., Yuan, C., Shan, Y.: Cobra: Efficient line art colorization with broader references. In: Proceedings of the Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers. pp. 1–11 (2025)

    2025

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.