Pith. sign in

REVIEW 4 major objections 6 minor 87 references

Multi-Task Label Discovery via Hierarchical Task Tokens for Partially Annotated Dense Predictions

T0 review · 4 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read The paper claims that learning compact hierarchical task tokens alongside a multi-task dense prediction network can discover pixel-level pseudo-supervision for tasks without annotations, and that this sets state-of-the-art results on…

desk verdict Genuinely new token-based pseudo-labeling method with strong raw results, but the headline ΔMTL numbers are not reproducible as reported because the metric convention is never fixed. read the letter →

arxiv 2411.18823 v2 pith:WS4Z5CO7 submitted 2024-11-27 cs.CV

Jingdong Zhang , Hanrong Ye , Xin Li , Wenping Wang , Dan Xu This is my paper · ORCID
classification cs.CV
keywords multi-tasklearningpartiallyannotateddensepredictionlabeldiscoveryhierarchicaltasktokenspseudo-labelingfeature-spacesupervisionprediction-spacecross-taskattention
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 is trying to establish that in partially annotated multi-task dense prediction — where each image is labeled for only some of the tasks it should predict — a set of compact learnable task tokens can recover the missing pixel-level supervision without heavy auxiliary mapping networks or diffusion decoders. The method learns a two-level token hierarchy: global tokens capture cross-task relations and provide feature-level pseudo-supervision, and fine-grained tokens inherited from them generate pixelwise pseudo-labels in prediction space. On PASCAL-Context, NYUD-v2, and Cityscapes, this yields state-of-the-art results under both one-label and random-labels annotation settings, including a 6.45-point gain in ΔMTL over XTC on NYUD-v2 one-label and an 8.76-point gain on Cityscapes, while using substantially fewer parameters than the best prior method.

What carries the argument

The load-bearing object is the two-level token hierarchy trained inside the decoder. In the global stage, tokens $\theta_i$ are updated by inter-task attention, which computes an affinity matrix $A = \mathrm{Softmax}(QK^\top) \in \mathbb{R}^{T\times T}$ and applies affine combinations $\Theta' = A\Theta$, $F' = AF$, so supervised tasks' representations are mixed into unsupervised task features; intra-task self-attention then refines each token-feature pair. In the fine-grained stage, updated $\theta_i$ are projected by an MLP into spatial task tokens $\varphi_i$, with orthogonal embeddings added to the MLP input to keep the token rows distinguishable, and $\varphi_i$ modulates the projected feature map through $G'_i = \mathrm{Conv}_{3\times3}(G_i \odot \mathrm{Softplus}(\varphi_i \times G_i))$. This hierarchy produces two losses: feature-level $L_f$ masked by confidence $M^f_i = \mathrm{Sigmoid}(\theta_i^\top F_i)$, and prediction-level $L_p$ using pseudo-labels $\tilde{Y}_i = \arg\max(\mathrm{Softmax}(G'_i))$ masked by $M^p_i = \max(\mathrm{Softmax}(G'_i)) > \tau_i$, summed as $L = L_s + L_p + L_f$.

What would settle it

A direct test would be to add a deliberately unrelated supervised task, such as predicting a fixed random class id per pixel, and compare the unlabeled tasks' accuracy with inter-task mixing enabled versus disabled; if the unrelated labels do not help or actively hurt, the cross-task affinity assumption is not transferring useful supervision.

Watch

Extended reading notes

Core claim

The central claim is that two hierarchies of tokens can discover consistent pixel-wise supervision for tasks that lack labels: global task tokens θ_i learn through inter- and intra-task attention over the shared feature map and yield feature-level pseudo-supervision, while fine-grained task tokens φ_i, projected from θ_i and regularized by orthogonal embeddings, refine task predictions and supply high-confidence pseudo-labels. The paper argues that because encoder features are shared across tasks, the learned tokens transfer discriminative information from labeled tasks to unlabeled ones, and that this transfer happens more cheaply and effectively than cross-task mapping networks or diffusion decoding. Its evidence is the reported performance on three benchmarks: on NYUD-v2 one-label, the method reaches 16.91% ΔMTL versus 10.46% for XTC, and on Cityscapes one-label it reaches 5.31% ΔMTL versus −3.45% for XTC; on PASCAL-Context it exceeds DiffusionMTL (F) by 2.44% ΔMTL with roughly 45% of its parameters and 70% of its FLOPs.

Load-bearing premise

The assumption that carries the method is that features from labeled tasks are informative enough to improve features of tasks without labels when mixed through the affinity matrix; if a supervised task is weakly related to an unsupervised one, the affine combination can inject noise instead of useful supervision.

Editorial extensions

If this is right

  • On NYUD-v2 one-label, the method's 16.91% ΔMTL (versus 10.46% for XTC) means partial annotations can support strong semantic segmentation, depth, and normal estimation simultaneously, so the limit is not the missing labels but how well the model can mine them.
  • On Cityscapes one-label, the method reaches 5.31% ΔMTL while XTC sits at −3.45%, and it is the only compared method with balanced gains on segmentation and depth, indicating that cross-task token transfer can turn an annotation imbalance into a symmetric improvement.
  • The ablations show feature-space and prediction-space supervision are complementary: $L_f$ alone adds 7.27 ΔMTL over baseline, $L_p$ alone adds 9.88, and together they add 11.30, implying both hierarchies should be trained jointly rather than choosing one.
  • On PASCAL-Context, the method improves ΔMTL by 2.44 points over DiffusionMTL (F) while using about 45% of its parameters and 70% of its GFLOPs, suggesting label discovery via tokens is a lighter alternative to diffusion-based pseudo-labeling.

Reading between the lines

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

  • Inference: if the inter-task affinity is the cause of the gains, then performance should scale with the number of labeled tasks available to mix and degrade when only one task carries ground truth; a controlled sweep over the number of labeled tasks would test this directly.
  • Inference: the same token hierarchy should transfer to other missing-supervision settings, such as sparse point labels, incomplete object masks, or absent input modalities, because the tokens learn a task's expected feature distribution rather than an explicit mapping from another task's predictions.
  • Inference: the orthogonality regularization suggests a boundary condition — when a task's categories are not separable by orthogonal directions, the fine-grained tokens may collapse and the pseudo-labels would need a different prior; testing on an ordinal regression task without binning would reveal this.
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 paper proposes Hierarchical Task Tokens (HiTTs), consisting of global task tokens θ_i and fine-grained task tokens φ_i, for partially annotated multi-task dense prediction. Global tokens are learned via inter- and intra-task attention and are used to discover feature-level pseudo supervision (L_f), while fine-grained tokens are projected from the global tokens with orthogonal embeddings and are used to discover prediction-level pseudo labels (L_p). The method follows a two-stage self-training pipeline: first train on labeled data, then generate offline pseudo supervision, then retrain from scratch. Experiments are reported on PASCAL-Context, NYUD-v2, and Cityscapes under one-label and random-labels settings, with claims of state-of-the-art ΔMTL gains over XTC and DiffusionMTL. Code is released.

Significance. If the quantitative claims are reproducible, the paper makes a significant contribution: it replaces heavy cross-task mapping networks with compact learnable tokens, reports lower parameter counts (62M vs. 133M) and GFLOPs than DiffusionMTL, and demonstrates consistent improvements across three benchmarks and multiple dense prediction tasks. The component ablations broadly support the design choices, and the release of code is a strength. However, the central ΔMTL claims are currently not uniquely determined because the main text and the supplementary material use different metric sets in Eq. (16) without stating the convention; until the tables are recomputed under a single convention, the exact magnitude of the claimed improvements cannot be verified.

major comments (4)
  1. [Section 4.2, Supp B.5.1, Eq. (16)] The ΔMTL values for identical raw metrics differ between the main text and the supplementary material. For NYUD-v2 one-label, Table 2 reports Ours=16.91 and XTC=10.46, while Table 6 reports Ours=13.23 and XTC=3.60 with the same Semseg/Depth/Normal primary metrics. For Cityscapes one-label, Table 3 reports Ours=5.31 and XTC=-3.45, while Table 8 reports Ours=3.45 and XTC=-3.53. The main text claims +6.45/+7.41/+8.76 ΔMTL over XTC, but Supp B.5.1 claims +9.63/+8.62/+6.98. Eq. (16) sums over N without ever fixing N or listing which metrics enter the sum. This makes the headline SOTA gains unverifiable. The authors must fix a single metric convention, state it explicitly, and recompute all ΔMTL values in both the main text and the supplement.
  2. [Section 4.3, Supp B.5.2, Table 4 vs. Table 9] The ablation conclusions also shift under the two ΔMTL conventions. The additive gain of HiTTs over the MTL baseline is +4.37 in Table 4 but +6.40 in Table 9, while the text in Supp B.5.2 says +9.64, which matches neither. The degradation from removing orthogonal embeddings is -7.32 in Table 4 but -4.38 in Table 9. Since every component comparison and the interpretation of which design choice matters is expressed in ΔMTL, the relative importance of the components is currently not reproducible. The same fix as above is required.
  3. [Section 3.1, Eqs. (5)-(6)] The cross-task affinity mechanism Θ' = A × Θ and F' = A × F assumes that affine combination with other task features improves the features of an unlabeled task. This is a load-bearing assumption for the proposed feature-supervision transfer, but the paper does not analyze the learned affinity matrix A or test settings with weakly correlated task pairs. If a labeled task is weakly related to an unlabeled task, the affine combination could inject noise rather than supervision. Add an analysis of the learned A and/or a sensitivity experiment with deliberately mismatched task subsets to demonstrate that the mechanism is robust.
  4. [All experimental tables] All reported results are from single runs without error bars or repeated seeds. Given the ΔMTL inconsistencies above, the absence of variance information makes it impossible to assess whether smaller margins (e.g., the PASCAL-Context gains of +2.44 and +1.46 over DiffusionMTL(F) in Table 1) are significant. Reporting mean and standard deviation over at least three seeds would materially strengthen the empirical claims.
minor comments (6)
  1. [Section 3.3, Eq. (2)] There is a duplicated word in the sentence introducing Eq. (2): “where where L_i(·) is the loss function.”
  2. [Section 4.2, NYUD-v2 paragraph] The text refers to “MTAN [28]” twice, but the MTAN reference is [33] in the bibliography; the citation should be corrected.
  3. [Table 1] The table uses “%” symbols in place of the checkmark/cross markers for model components (e.g., “MTDNet” and “Mapping Network”), which is confusing and should be replaced with readable symbols or –/✓ entries.
  4. [Table 13] The header lists both Depth and Normal columns as “mErr↓”; the Depth column should be an absolute error metric (AbS or absErr) consistent with the other tables.
  5. [Figure 14] The caption reads “Quantitative analysis of the quality of pseudo labels,” but the figure shows qualitative visualizations; the caption should be corrected to “qualitative comparison.”
  6. [Section 3.3] The sentence “each item is combined with the weight 1 to form the total loss” is ambiguous; specify whether all three loss terms are simply added with equal weight and whether any loss balancing is applied.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the method's pseudo-labeling loop is self-training, not a derivation that reduces to its inputs; the metric inconsistency in ΔMTL is a verifiability defect, not a circular step.

full rationale

The paper's chain is empirical: Eqs. 2-4 define losses on ground-truth labels, pseudo-labels, and saved features, and Eq. 16 defines the aggregate evaluation metric. The pseudo-labels and feature targets are produced offline by the model's own frozen weights after training on labeled data (Sec. 4.1, Training Pipeline: 'We produce the pseudo label in an offline manner according to [65]... train the multi-task model from scratch'), which is standard self-training and is not a derivation whose output is equivalent to its input. Final claims are evaluated on fixed test sets against external baselines (XTC, MTAN, DiffusionMTL), so the reported gains are not forced by construction. The paper imports no uniqueness theorem, and its self-citations ([28], [71], [73], [74]) are baselines or related work, not load-bearing premises. One non-circular defect is flagged per the review rule: Eq. 16 defines ΔMTL as a sum over N, but N is never fixed. Main Table 2 (NYUD-v2 one-label) reports MTLbaseline/XTC/Ours ΔMTL = 5.61/10.46/16.91 using only Semseg mIoU, Depth AbS, and Normal mErr, while supplement Table 6, with identical raw numbers for those three metrics, reports 0.11/3.60/13.23 using the full 9-metric set; Cityscapes Ours is 5.31 in Table 3 but 3.45 in Table 8. The exact headline gaps (+6.45, +7.41, +8.76) therefore depend on an unstated metric-choice convention and cannot be uniquely recomputed until the convention is fixed. This is a reproducibility/verifiability defect, not circularity; the qualitative direction of improvement is consistent across both conventions.

Assumptions & free parameters 3 free parameters · 4 assumptions · 2 invented entities

The method adds trained token parameters, hand-picked confidence thresholds, and a self-training loop; the central improvements rest on these choices.

free parameters (3)
  • per-task confidence thresholds τ_i = NYUD-v2 one-label: Semseg 0.9, Depth 0.45, Normal 0.6; other datasets/settings in supplementary B.3
    Chosen by hand per dataset/task to mask pseudo labels; central to L_p quality.
  • number of depth bins / normal cluster centers = NYUD-v2 depth 30, normal 20; Cityscapes depth 100; PASCAL-Context normal 40
    Discrete quantization of regression tasks, selected following prior settings; affects fine-grained token supervision.
  • global/fine task token dimensions and training hyperparameters = C, C_p, LR 2e-5/1e-4, epochs 100/400, etc.
    Standard network hyperparameters, not fitted to target but required to reproduce.
assumptions (4)
  • ad hoc to paper Cross-task affinity matrix A can transfer discriminative representations from supervised to unsupervised task features.
    Eq. 6, Sec 3.1. If tasks are weakly correlated, affine combination can inject noise.
  • domain assumption The pretrained model's own predictions on unlabeled tasks are reliable enough to serve as pseudo labels when confidence exceeds τ_i.
    Sec 3.3, Eq. 4. Self-training assumption; no mechanism prevents confirmation bias.
  • domain assumption Discrete quantization of regression labels (following [4]) does not change task learning and preserves fair comparison.
    Sec 3, A.3 cites [4]; if quantization hurts regression tasks, comparisons are affected.
  • ad hoc to paper The offline pseudo feature targets F̃_i saved from the pretrained model remain valid after retraining from scratch.
    Eq. 3 in Sec 3.3; feature MSE against a fixed snapshot can drift as features change.
invented entities (2)
  • Global task tokens θ_i
    purpose: Learned per-task query vectors for cross-task attention and feature supervision discovery.
    Internal model parameters; no falsifiable prediction outside the paper.
  • Fine-grained task tokens φ_i
    purpose: Per-category spatial tokens derived from θ_i via MLP with orthogonal embeddings, refine task features and pseudo labels.
    Internal model parameters; the orthogonal-embedding prior is a design choice.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multi-Task Label Discovery via Hierarchical Task Tokens for Partially Annotated Dense Predictions." pith.science (2026). https://pith.science/paper/WS4Z5CO7

@misc{pith2026241118823,
  author       = {Pith},
  title        = {Pith review of: Multi-Task Label Discovery via Hierarchical Task Tokens for Partially Annotated Dense Predictions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WS4Z5CO7}},
  note         = {Machine review of arXiv:2411.18823}
}
read the original abstract

In recent years, simultaneous learning of multiple dense prediction tasks with partially annotated label data has emerged as an important research area. Previous works primarily focus on leveraging cross-task relations or conducting adversarial training for extra regularization, which achieve promising performance improvements, while still suffering from the lack of direct pixel-wise supervision and extra training of heavy mapping networks. To effectively tackle this challenge, we propose a novel approach to optimize a set of compact learnable hierarchical task tokens, including global and fine-grained ones, to discover consistent pixel-wise supervision signals in both feature and prediction levels. Specifically, the global task tokens are designed for effective cross-task feature interactions in a global context. Then, a group of fine-grained task-specific spatial tokens for each task is learned from the corresponding global task tokens. It is embedded to have dense interactions with each task-specific feature map. The learned global and local fine-grained task tokens are further used to discover pseudo task-specific dense labels at different levels of granularity, and they can be utilized to directly supervise the learning of the multi-task dense prediction framework. Extensive experimental results on challenging NYUD-v2, Cityscapes, and PASCAL Context datasets demonstrate significant improvements over existing state-of-the-art methods for partially annotated multi-task dense prediction.

Figures

Figures reproduced from arXiv: 2411.18823 by the authors.

Figure 1
Figure 1. (a) Illustration of partially annotated multi-task dense prediction setting. Each input image only has partial task labels [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Illustration of our method. HiTTs consist of both global and fine-grained task tokens which learn discriminative task [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. (a) Global Task Token Learning. It mainly contains two stages: Inter- and Intra-Task Attention for thoroughly [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: Illustration of Token & Feature Projection and the Fine-grained Task Token Learning. Different tasks have the same structure, and we take one as an example. We project the fine-grained feature 𝑮𝒊 from the updated task feature 𝑭𝒊 , and derive fine-grained task tokens 𝝋𝒊…
Figure 5
Figure 5. Figure 5: Comparisons with SoTA works on NYUDv2. Ours [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Comparisons of task score maps produced by global task tokens and fine-grained task tokens on Cityscapes. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Comparison of the training and testing performance [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Illustration of detail designs of: (i) Inter-Task Attention: predicting cross-task token affinities [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: Illustrations of Single-scale Global Task Token Learning (a) and Multi-scale Global Task Token Learning (b). For the multi-scale features produced by the shared backbone, we use linear layers to produce corresponding task tokens for each scale. Then, in each scale, we …
Figure 10
Figure 10. Figure 10: Comparison of the task confidence map before and [PITH_FULL_IMAGE:figures/full_fig_p017_10.png]
Figure 11
Figure 11. Figure 11: Visualization of self-affinities heatmap ( [PITH_FULL_IMAGE:figures/full_fig_p019_11.png]
Figure 12
Figure 12. Figure 12: Comparisons of task score maps produced by global task tokens and fine-grained task tokens. The upper part is the [PITH_FULL_IMAGE:figures/full_fig_p020_12.png]
Figure 13
Figure 13. Figure 13: Comparisons with SOTA works on NYUD-v2 (upper part) and Cityscapes (lower part). [PITH_FULL_IMAGE:figures/full_fig_p021_13.png]
Figure 14
Figure 14. Figure 14: Quantitative analysis of the quality of pseudo labels generated by gloabl task tokens. [PITH_FULL_IMAGE:figures/full_fig_p022_14.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

87 extracted references · 57 canonical work pages

  1. [1]

    Vijay Badrinarayanan, Alex Kendall, and Roberto Cipolla. 2017. Segnet: A deep convolutional encoder-decoder architecture for image segmentation. PAMI 39, 12 (2017), 2481–2495

    2017

  2. [2]

    Jinze Bai, Shuai Bai, Shusheng Yang, Shijie Wang, Sinan Tan, Peng Wang, Junyang Lin, Chang Zhou, and Jingren Zhou. 2023. Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyond.arXiv preprint arXiv:2308.12966 (2023)

    arXiv 2023

  3. [3]

    Alexey Bochkovskiy, Chien-Yao Wang, and Hong-Yuan Mark Liao. 2020. Yolov4: Optimal speed and accuracy of object detection. arXiv preprint arXiv:2004.10934 (2020)

    arXiv 2020

  4. [4]

    David Brüggemann, Menelaos Kanakis, Anton Obukhov, Stamatios Georgoulis, and Luc Van Gool. 2021. Exploring relational context for multi-task dense pre- diction. In ICCV. 15869–15878

    2021

  5. [5]

    Yancheng Cai, Bo Zhang, Baopu Li, Tao Chen, Hongliang Yan, Jingdong Zhang, and Jiahao Xu. 2023. Rethinking cross-domain pedestrian detection: A background-focused distribution alignment framework for instance-free one- stage detectors. IEEE transactions on image processing 32 (2023), 4935–4950

    2023

  6. [6]

    Mang Cao, Sanping Zhou, Ye Deng, Wenli Huang, Le Wang, and Jinjun Wang. [n. d.]. MSM: Multi-Scale Mamba in Multi-Task Dense Prediction. ([n. d.])

  7. [7]

    Ruchika Chavhan, Abhinav Mehrotra, Malcolm Chadwick, Alberto Gil Ramos, Luca Morreale, Mehdi Noroozi, and Sourav Bhattacharya. 2025. Upcycling Text-to-Image Diffusion Models for Multi-Task Capabilities. arXiv preprint arXiv:2503.11905 (2025)

    work page Pith review arXiv 2025

  8. [8]

    Linwei Chen, Ying Fu, Kaixuan Wei, Dezhi Zheng, and Felix Heide. 2023. Instance segmentation in the dark. International Journal of Computer Vision 131, 8 (2023), 2198–2218

    2023

Show all 87 references
  1. [9]

    Liang-Chieh Chen, Yukun Zhu, George Papandreou, Florian Schroff, and Hartwig Adam. 2018. Encoder-decoder with atrous separable convolution for semantic image segmentation. In Proceedings of the European conference on computer vision (ECCV). 801–818

    2018

  2. [10]

    Zhao Chen, Vijay Badrinarayanan, Chen-Yu Lee, and Andrew Rabinovich. 2018. Gradnorm: Gradient normalization for adaptive loss balancing in deep multitask networks. In International conference on machine learning . PMLR, 794–803

    2018

  3. [11]

    Bowen Cheng, Ishan Misra, Alexander G Schwing, Alexander Kirillov, and Rohit Girdhar. 2022. Masked-attention mask transformer for universal image segmen- tation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 1290–1299

    2022

  4. [12]

    Marius Cordts, Mohamed Omran, Sebastian Ramos, Timo Rehfeld, Markus En- zweiler, Rodrigo Benenson, Uwe Franke, Stefan Roth, and Bernt Schiele. 2016. The cityscapes dataset for semantic urban scene understanding. In CVPR. 3213–3223. doi:10.1109/CVPR.2016.350

    2016 doi

  5. [13]

    Camille Couprie, Clément Farabet, Laurent Najman, and Yann LeCun. 2013. Indoor semantic segmentation using depth information. arXiv preprint arXiv:1301.3572 (2013)

    2013 arXiv

  6. [14]

    Jean-Antoine Désidéri. 2012. Multiple-gradient descent algorithm (MGDA) for multiobjective optimization. Comptes Rendus Mathematique 350, 5-6 (2012), 313–318

    2012

  7. [15]

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xi- aohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. 2020. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv prepri...

    2020 arXiv

  8. [16]

    David Eigen and Rob Fergus. 2015. Predicting depth, surface normals and se- mantic labels with a common multi-scale convolutional architecture. In ICCV. 2650–2658

    2015

  9. [17]

    Mark Everingham, Luc Van Gool, Christopher KI Williams, John Winn, and Andrew Zisserman. 2009. The pascal visual object classes (voc) challenge. IJCV 88 (2009), 303–308. doi:10.1007/s11263-009-0275-4

    2009 doi

  10. [18]

    Yuan Gao, Jiayi Ma, Mingbo Zhao, Wei Liu, and Alan L Yuille. 2019. Nddr-cnn: Layerwise feature fusing in multi-task cnns by neural discriminative dimension- ality reduction. In CVPR. 3205–3214

    2019

  11. [19]

    Zhangxuan Gu, Haoxing Chen, and Zhuoer Xu. 2024. Diffusioninst: Diffusion model for instance segmentation. In ICASSP 2024-2024 IEEE International Confer- ence on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2730–2734

    2024

  12. [20]

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep residual learning for image recognition. In CVPR. 770–778

    2016

  13. [21]

    Lukas Hoyer, David Joseph Tan, Muhammad Ferjad Naeem, Luc Van Gool, and Federico Tombari. 2024. SemiVL: semi-supervised semantic segmentation with vision-language guidance. In European Conference on Computer Vision . Springer, 257–275

    2024

  14. [22]

    Abdullah-Al-Zubaer Imran, Chao Huang, Hui Tang, Wei Fan, Yuan Xiao, Dingjun Hao, Zhen Qian, and Demetri Terzopoulos. 2020. Partly Supervised Multitask Learning. arXiv preprint arXiv:2005.02523 (2020)

    2020 arXiv

  15. [23]

    Ahmet Iscen, Giorgos Tolias, Yannis Avrithis, and Ondrej Chum. 2019. Label propagation for deep semi-supervised learning. In CVPR. 5070–5079

    2019

  16. [24]

    Maximilian Jaritz, Jiayuan Gu, and Hao Su. 2019. Multi-view pointnet for 3d scene understanding. In Proceedings of the IEEE/CVF international conference on computer vision workshops. 0–0

    2019

  17. [25]

    Alex Kendall, Yarin Gal, and Roberto Cipolla. 2018. Multi-task learning using uncertainty to weigh losses for scene geometry and semantics. In Proceedings of the IEEE conference on computer vision and pattern recognition . 7482–7491

    2018

  18. [26]

    Dong-Hyun Lee et al . 2013. Pseudo-label: The simple and efficient semi- supervised learning method for deep neural networks. In ICML, Vol. 3. 896

    2013

  19. [27]

    Bo Li, Yuchao Dai, and Mingyi He. 2018. Monocular depth estimation with hierarchical fusion of dilated cnns and soft-weighted-sum inference. Pattern Recognition 83 (2018), 328–339

    2018

  20. [28]

    Wei-Hong Li, Xialei Liu, and Hakan Bilen. 2022. Learning multiple dense predic- tion tasks from partially annotated data. In CVPR. 18879–18889

    2022

  21. [29]

    Yunsheng Li, Lu Yuan, and Nuno Vasconcelos. 2019. Bidirectional learning for domain adaptation of semantic segmentation. In CVPR. 6936–6945

    2019

  22. [30]

    Bin Lin, Yang Ye, Bin Zhu, Jiaxi Cui, Munan Ning, Peng Jin, and Li Yuan. 2023. Video-llava: Learning united visual representation by alignment before projection. arXiv preprint arXiv:2311.10122 (2023)

    2023 arXiv

  23. [31]

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. 2023. Visual in- struction tuning. Advances in neural information processing systems 36 (2023), 34892–34916

    2023

  24. [32]

    Qiuhua Liu, Xuejun Liao, and Lawrence Carin. 2007. Semi-supervised multitask learning. NIPS 20 (2007)

    2007

  25. [33]

    Shikun Liu, Edward Johns, and Andrew J Davison. 2019. End-to-end multi-task learning with attention. In CVPR. 1871–1880

    2019

  26. [34]

    Yuxiang Lu, Shengcao Cao, and Yu-Xiong Wang. 2024. Swiss army knife: Syner- gizing biases in knowledge from vision foundation models for multi-task learning. arXiv preprint arXiv:2410.14633 (2024)

    2024 arXiv

  27. [35]

    Yao Lu, Soren Pirk, Jan Dlabal, Anthony Brohan, Ankita Pasad, Zhao Chen, Vincent Casser, Anelia Angelova, and Ariel Gordon. 2021. Taskology: Utilizing task relations at scale. In CVPR. 8700–8709

    2021

  28. [36]

    Xiangde Luo, Jieneng Chen, Tao Song, and Guotai Wang. 2021. Semi-supervised medical image segmentation through dual-task consistency. In AAAI, Vol. 35. 8801–8809

    2021

  29. [37]

    Huayu Mai, Rui Sun, Tianzhu Zhang, and Feng Wu. 2024. Rankmatch: Exploring the better consistency regularization for semi-supervised semantic segmenta- tion. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 3391–3401

    2024

  30. [38]

    Kevis-Kokitsi Maninis, Ilija Radosavovic, and Iasonas Kokkinos. 2019. Attentive single-tasking of multiple tasks. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . 1851–1860

    2019

  31. [39]

    David R Martin, Charless C Fowlkes, and Jitendra Malik. 2004. Learning to detect natural image boundaries using local brightness, color, and texture cues. PAMI 26, 5 (2004), 530–549

    2004

  32. [40]

    Ishan Misra, Abhinav Shrivastava, Abhinav Gupta, and Martial Hebert. 2016. Cross-stitch networks for multi-task learning. In CVPR. 3994–4003

    2016

  33. [41]

    Kento Nishi, Junsik Kim, Wanhua Li, and Hanspeter Pfister. 2024. Joint-Task Regularization for Partially Labeled Multi-Task Learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 16152–16162

    2024

  34. [42]

    Songyou Peng, Kyle Genova, Chiyu Jiang, Andrea Tagliasacchi, Marc Pollefeys, Thomas Funkhouser, et al. 2023. Openscene: 3d scene understanding with open vocabularies. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 815–824

    2023

  35. [43]

    Alessandro Pieropan, Hossein Azizpour, Atsuto Maki, et al. 2022. Dense FixMatch: a simple semi-supervised learning method for pixel-wise prediction tasks. arXiv preprint arXiv:2210.09919 (2022)

    2022 arXiv

  36. [44]

    Lingyan Ran, Yali Li, Guoqiang Liang, and Yanning Zhang. 2024. Pseudo label- ing methods for semi-supervised semantic segmentation: A review and future perspectives. IEEE Transactions on Circuits and Systems for Video Technology (2024)

    2024

  37. [45]

    Tianhe Ren, Yihao Chen, Qing Jiang, Zhaoyang Zeng, Yuda Xiong, Wenlong Liu, Zhengyu Ma, Junyi Shen, Yuan Gao, Xiaoke Jiang, et al. 2024. Dino-x: A unified vision model for open-world object detection and understanding. arXiv preprint arXiv:2411.14347 (2024)

    2024 arXiv

  38. [46]

    Jonas Schult, Francis Engelmann, Alexander Hermans, Or Litany, Siyu Tang, and Bastian Leibe. 2022. Mask3d: Mask transformer for 3d semantic instance segmentation. arXiv preprint arXiv:2210.03105 (2022)

    2022 arXiv

  39. [47]

    Weiwei Shi, Yihong Gong, Chris Ding, Zhiheng MaXiaoyu Tao, and Nanning Zheng. 2018. Transductive semi-supervised deep learning using min-max features. In ECCV. 299–315

    2018

  40. [48]

    Nathan Silberman, Derek Hoiem, Pushmeet Kohli, and Rob Fergus. 2012. Indoor segmentation and support inference from rgbd images. In ECCV. Springer, 746–

    2012

  41. [49]

    Kihyuk Sohn, David Berthelot, Nicholas Carlini, Zizhao Zhang, Han Zhang, Colin A Raffel, Ekin Dogus Cubuk, Alexey Kurakin, and Chun-Liang Li. 2020. Fixmatch: Simplifying semi-supervised learning with consistency and confidence. NIPS 33 (2020), 596–608

    2020

  42. [50]

    Roger Alan Stein, Patricia A Jaques, and Joao Francisco Valiati. 2019. An analysis of hierarchical text classification using word embeddings. Information Sciences 471 (2019), 216–232. MM ’25, October 27–31, 2025, Dublin, Ireland Zhang et al

    2019

  43. [51]

    Aixin Sun and Ee-Peng Lim. 2001. Hierarchical text classification and evaluation. In Proceedings 2001 IEEE International Conference on Data Mining . IEEE, 521–528

    2001

  44. [52]

    Hui Tang and Kui Jia. 2022. Towards Discovering the Effectiveness of Moderately Confident Samples for Semi-Supervised Learning. In CVPR. 14658–14667

    2022

  45. [53]

    Yingjie Tang, Shou Feng, Chunhui Zhao, Yongqi Chen, Zhiyong Lv, and Weiwei Sun. 2025. A Semantic Change Detection Network Based on Boundary Detec- tion and Task Interaction for High-Resolution Remote Sensing Images. IEEE Transactions on Neural Networks and Learning Systems (2025)

    2025

  46. [54]

    Antti Tarvainen and Harri Valpola. 2017. Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results. NIPS 30 (2017)

    2017

  47. [55]

    Yuxin Tian, Yijie Lin, Qing Ye, Jian Wang, Xi Peng, and Jiancheng Lv. 2024. UNITE: multitask learning with sufficient feature for dense prediction. IEEE Transactions on Systems, Man, and Cybernetics: Systems 54, 8 (2024), 5012–5024

    2024

  48. [56]

    Simon Vandenhende, Stamatios Georgoulis, Wouter Van Gansbeke, Marc Proes- mans, Dengxin Dai, and Luc Van Gool. 2021. Multi-task learning for dense prediction tasks: A survey. PAMI (2021)

    2021

  49. [57]

    Simon Vandenhende, Stamatios Georgoulis, and Luc Van Gool. 2020. Mti-net: Multi-scale task interaction networks for multi-task learning. In ECCV. Springer, 527–543

    2020

  50. [58]

    Yizhou Wang, Kuan-Chuan Peng, and Yun Fu. 2025. Towards zero-shot 3d anomaly localization. In 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (W ACV). IEEE, 1447–1456

    2025

  51. [59]

    Yizhou Wang, Can Qin, Yue Bai, Yi Xu, Xu Ma, and Yun Fu. 2022. Making reconstruction-based method great again for video anomaly detection. In 2022 IEEE International Conference on Data Mining (ICDM) . IEEE, 1215–1220

    2022

  52. [60]

    Yizhou Wang, Can Qin, Rongzhe Wei, Yi Xu, Yue Bai, and Yun Fu. 2022. Self- supervision meets adversarial perturbation: A novel framework for anomaly detection. In Proceedings of the 31st ACM International Conference on Information & Knowledge Management. 4555–4559

    2022

  53. [61]

    Yizhou Wang, Can Qin, Rongzhe Wei, Yi Xu, Yue Bai, and Yun Fu. 2024. Sla ^2 p: Self-supervised anomaly detection with adversarial perturbation. IEEE Transactions on Knowledge and Data Engineering (2024)

    2024

  54. [62]

    Yufeng Wang, Yi-Hsuan Tsai, Wei-Chih Hung, Wenrui Ding, Shuo Liu, and Ming- Hsuan Yang. 2022. Semi-supervised multi-task learning for semantics and depth. In W ACV. 2505–2514

    2022

  55. [63]

    Yizhou Wang, Lingzhi Zhang, Yue Bai, Mang Tik Chiu, Zhengmian Hu, Mingyuan Zhang, Qihua Dong, Yu Yin, Sohrab Amirghodsi, and Yun Fu. 2025. Cautious Next Token Prediction. arXiv preprint arXiv:2507.03038 (2025)

    2025 arXiv

  56. [64]

    Yizhou Wang, Ruiyi Zhang, Haoliang Wang, Uttaran Bhattacharya, Yun Fu, and Gang Wu. 2023. Vaquita: Enhancing alignment in llm-assisted video understand- ing. arXiv preprint arXiv:2312.02310 (2023)

    2023 arXiv

  57. [65]

    Qizhe Xie, Minh-Thang Luong, Eduard Hovy, and Quoc V Le. 2020. Self- training with noisy student improves imagenet classification. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . 10687–10698

    2020

  58. [66]

    Dan Xu, Wanli Ouyang, Xiaogang Wang, and Nicu Sebe. 2018. Pad-net: Multi- tasks guided prediction-and-distillation network for simultaneous depth estima- tion and scene parsing. In CVPR. 675–684

    2018

  59. [67]

    Lihe Yang, Lei Qi, Litong Feng, Wayne Zhang, and Yinghuan Shi. 2023. Revisit- ing weak-to-strong consistency in semi-supervised semantic segmentation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . 7236–7246

    2023

  60. [68]

    Lihe Yang, Zhen Zhao, and Hengshuang Zhao. 2025. Unimatch v2: Pushing the limit of semi-supervised semantic segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence (2025)

    2025

  61. [69]

    Siwei Yang, Hanrong Ye, and Dan Xu. 2023. Contrastive multi-task dense predic- tion. In Proceedings of the AAAI Conference on Artificial Intelligence (AAAI)

    2023

  62. [70]

    Yuqi Yang, Peng-Tao Jiang, Qibin Hou, Hao Zhang, Jinwei Chen, and Bo Li. [n. d.]. Multi-Task Dense Predictions via Unleashing the Power of Diffusion. In The Thirteenth International Conference on Learning Representations

  63. [71]

    Hanrong Ye and Dan Xu. 2022. Inverted Pyramid Multi-task Transformer for Dense Scene Understanding. ECCV (2022)

    2022

  64. [72]

    Hanrong Ye and Dan Xu. 2023. TaskExpert: Dynamically Assembling Multi- Task Representations with Memorial Mixture-of-Experts. In Proceedings of the IEEE/CVF International Conference on Computer Vision . 21828–21837

    2023

  65. [73]

    Hanrong Ye and Dan Xu. 2023. TaskPrompter: Spatial-Channel Multi-Task Prompting for Dense Scene Understanding. In ICLR

    2023

  66. [74]

    Hanrong Ye and Dan Xu. 2024. DiffusionMTL: Learning Multi-Task Denoising Diffusion Model from Partially Annotated Data. In CVPR

    2024

  67. [75]

    Amir R Zamir, Alexander Sax, Nikhil Cheerla, Rohan Suri, Zhangjie Cao, Jiten- dra Malik, and Leonidas J Guibas. 2020. Robust learning through cross-task consistency. In CVPR. 11197–11206

    2020

  68. [76]

    Bernhard Zeisl, Marc Pollefeys, et al. 2014. Discriminatively trained dense surface normal estimation. In ECCV. Springer, 468–484

    2014

  69. [77]

    Yu Zeng, Yunzhi Zhuge, Huchuan Lu, and Lihe Zhang. 2019. Joint learning of saliency detection and weakly supervised semantic segmentation. In ICCV. 7223–7233

    2019

  70. [78]

    Jingdong Zhang, Jiayuan Fan, Peng Ye, Bo Zhang, Hancheng Ye, Baopu Li, Yancheng Cai, and Tao Chen. 2023. Rethinking of Feature Interaction for Multi- task Learning on Dense Prediction. arXiv preprint arXiv:2312.13514 (2023)

    2023 arXiv

  71. [79]

    Jingdong Zhang, Jiayuan Fan, Peng Ye, Bo Zhang, Hancheng Ye, Baopu Li, Yancheng Cai, and Tao Chen. 2025. BridgeNet: Comprehensive and Effective Feature Interactions via Bridge Feature for Multi-Task Dense Predictions. IEEE Transactions on Pattern Analysis and Machine Intellige...

    2025

  72. [80]

    Zhenyu Zhang, Zhen Cui, Chunyan Xu, Zequn Jie, Xiang Li, and Jian Yang. 2018. Joint task-recursive learning for semantic segmentation and depth estimation. In ECCV. 235–251

    2018

  73. [81]

    Zhenyu Zhang, Zhen Cui, Chunyan Xu, Yan Yan, Nicu Sebe, and Jian Yang

  74. [82]

    Jie Zhou, Chunping Ma, Dingkun Long, Guangwei Xu, Ning Ding, Haoyu Zhang, Pengjun Xie, and Gongshen Liu. 2020. Hierarchy-aware global model for hi- erarchical text classification. In Proceedings of the 58th annual meeting of the association for computational linguistics . 1106–1117

    2020

  75. [83]

    Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, and Jifeng Dai. 2020. Deformable detr: Deformable transformers for end-to-end object detection.arXiv preprint arXiv:2010.04159 (2020)

    2020 arXiv

  76. [84]

    Yang Zou, Zhiding Yu, BVK Kumar, and Jinsong Wang. 2018. Unsupervised domain adaptation for semantic segmentation via class-balanced self-training. In ECCV. 289–305

    2018

  77. [85]

    Yang Zou, Zhiding Yu, Xiaofeng Liu, BVK Kumar, and Jinsong Wang. 2019. Confidence regularized self-training. In ICCV. 5982–5991. Multi-Task Label Discovery via Hierarchical Task Tokens for Partially Annotated Dense Predictions MM ’25, October 27–31, 2025, Dublin, Ireland Suppl...

    2019

  78. [760]

    doi:10.1007/978-3-642-33715-4_54

    doi

  79. [2019]

    Pattern-affinitive propagation across depth, surface normal and semantic segmentation. In CVPR. 4106–4115

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.