Pith. sign in

REVIEW 5 major objections 5 minor 51 references

Multi-Task Dense Prediction Fine-Tuning with Mixture of Fine-Grained Experts

T0 review · 5 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A decoder built from fine-grained, shared, and global mixture-of-experts layers can beat full-model multi-task training while keeping the backbone frozen and training only 2.5–4.7M parameters.

desk verdict Plausible fine-grained MoE decoder idea, but the paper's own tables contradict its headline performance claims, so the evidence is not usable as reported. read the letter →

arxiv 2507.19077 v1 pith:JIG42SMC submitted 2025-07-25 cs.CV

Yangyang Xu , Xi Ye , Duo Su This is my paper · ORCID
classification cs.CV
keywords SceneUnderstandingMulti-taskLearningMixture-of-ExpertsFine-TuningDensePredictionParameter-EfficientSharedExpertsGlobalExpert
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

This paper claims that multi-task dense prediction—predicting segmentation, depth, normals, boundaries, and saliency from one image—can be done better by training only a small decoder on top of a frozen Swin Transformer than by fine-tuning the whole network. The proposed architecture, FGMoE, replaces monolithic task-specific decoders with three kinds of mixture-of-experts blocks: fine-grained intra-task experts, always-on shared experts, and a global expert layer for cross-task knowledge. On NYUD-v2 and PASCAL-Context, the authors report that FGMoE outperforms prior MoE-based multi-task models such as TaskExpert and MLoRE while using substantially fewer trainable parameters. If the claim holds, parameter-efficient fine-tuning is not a compromise but a route to stronger multi-task dense prediction.

What carries the argument

The load-bearing mechanism is the task-specific MoE block, whose output is the sum of always-on shared experts plus a top-k weighted combination of fine-grained intra-task experts. Intra-task experts are created by splitting the intermediate hidden dimension of an MLP, so more experts can be activated without adding parameters; shared experts, fixed at two, consolidate common context and reduce redundancy in the routed experts; and a global MoE layer, using the same block structure, aggregates knowledge across tasks and is added to each task-specific output before the prediction head. Around these blocks, a feature aggregator fuses four Swin stages and a deformable mixer, combining channel convolution, deformable convolution, and multi-head self-attention, aligns spatial features so the decoder can specialize per task while the backbone stays frozen.

What would settle it

Train the identical FGMoE decoder twice on NYUD-v2 with Swin-T, once with the backbone frozen and once with the final backbone stage unfrozen under matched parameter and compute budgets; if unfreezing raises $\Delta_m$ by more than a small margin, the claim that decoder-only fine-tuning is sufficient collapses, and if a decoder on a randomly initialized frozen backbone recovers comparable $\Delta_m$, the decoder architecture rather than the pretrained features would be doing the work.

Watch

Extended reading notes

Core claim

The paper claims that the routing granularity of a decoder, not its parameter count or backbone fine-tuning, is what determines multi-task dense-prediction quality. FGMoE splits each expert's hidden MLP dimension into fine-grained sub-experts so that a fixed parameter budget activates more experts, marks a subset of experts as always-active shared experts to absorb redundant common features, and adds a global MoE layer that routes cross-task information per pixel and per task. With the Swin backbone frozen and only the roughly 2.5–4.7M-parameter decoder trained, FGMoE reports higher mean per-task improvement than the leading decoder-only tuning baselines on PASCAL-Context and NYUD-v2, and with full training it beats TaskExpert and MLoRE on PASCAL-Context with a ViT-L backbone while using fewer decoder parameters. The intended upshot is that task-specific specialization and cross-task sharing can be engineered entirely in a small decoder, making parameter-efficient fine-tuning the preferred mode for multi-task dense prediction.

Load-bearing premise

The load-bearing premise is that a frozen Swin Transformer already contains all the generic visual features these dense tasks need, so no backbone parameter has to change; if the frozen features are insufficient for a task, the small decoder has no way to recover them.

Editorial extensions

If this is right

  • If FGMoE is right, decoder-only fine-tuning becomes the default mode for multi-task dense prediction: with 4.7M trainable parameters on PASCAL-Context with Swin-L it reaches +7.89% mean per-task improvement, slightly above DITASK's +7.79% at 7.13M parameters.
  • Task-specific decoders no longer need per-task parameter pools; fine-grained expert splitting lets a fixed decoder budget cover multiple tasks, so task count can grow without proportional decoder growth.
  • The global expert layer that routes per pixel and per task can reduce negative transfer: FGMoE reports a positive $\Delta_m$ on full PASCAL-Context training with ViT-L (+0.39%) where prior MoE decoders report negative drops, meaning multi-task training is no longer a loss relative to single-task baselines.
  • The ablation result that 8 routed experts outperform 16, and that top-k near 4–6 is best, implies expert count and sparsity should be tuned as hyperparameters rather than maximized; the paper itself lists adaptive top-k as future work.
  • A small positive $\Delta_m$ in decoder-only fine-tuning, despite using only about 2–4% of the parameters of full-network training, suggests parameter-efficient multi-task models can be deployed where full fine-tuning is too costly.

Reading between the lines

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

  • A natural but untested extension is to swap the frozen Swin backbone for other frozen feature extractors such as convnets or self-supervised ViTs; if the decoder is the true source of task specialization, the gains should transfer, and if not, the frozen-feature premise is what carries the result.
  • The fine-grained expert-splitting trick is parameter-agnostic: splitting MLP hidden dimensions could be applied to any MoE layer outside dense prediction, such as language or speech decoders, to raise expert count without raising parameter count; the paper does not claim this.
  • The small depth-error gap on NYUD-v2 (FGMoE rmse 0.618 vs DITASK 0.606) hints that the average $\Delta_m$ metric may hide per-task trade-offs; a metric-by-metric comparison rather than only the mean would clarify where the advantage actually lies.
  • If the frozen-backbone premise is pushed, one testable corollary is that FGMoE's advantage should shrink on tasks requiring high-frequency detail not present in Swin features, such as boundary detection at low resolution; a controlled backbone-capacity study would settle it.
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

5 major / 5 minor

Summary. The paper proposes FGMoE, a decoder architecture for multi-task dense prediction that combines three components: intra-task experts obtained by splitting the intermediate hidden dimensions of expert MLPs, always-active shared experts that consolidate common information, and a global MoE layer for cross-task knowledge transfer. The model is evaluated on NYUD-v2 and PASCAL-Context with several backbones, under both full-model training and decoder-only fine-tuning. The abstract claims that FGMoE uses fewer parameters than existing MoE-based MTL models and significantly outperforms them on both datasets. The paper also reports ablations on the proposed components, the number of experts, and the top-k routing value.

Significance. If the reported results were accurate, the proposed fine-grained expert decomposition and global cross-task routing would be a useful addition to parameter-efficient multi-task dense prediction. The ablation studies in Section 4.3 provide component-level evidence that the deformable mixer, task-specific MoEs, and global MoEs contribute positively, and the decoder-only fine-tuning results in Tables 3 and 4 indicate competitive parameter counts. However, the paper's central comparative claim is not supported by its own reported numbers: Table 2 shows FGMoE underperforming MLoRE on NYUD-v2, and Section 4.2's text describing HRNet18 results contradicts Table 1. These internal inconsistencies prevent the reader from assessing the actual significance of the method, despite the plausibility of the architectural ideas.

major comments (5)
  1. [Abstract and Table 2] The abstract claims FGMoE "significantly outperforms current MoE-based competitive MTL models," but Table 2 (NYUD-v2, ViT-L) shows FGMoE with surface normal mErr of 19.19 versus 18.33 for MLoRE and 18.56 for the single-task baseline, and with Δm of -0.58 versus +0.11 for MLoRE. Since MLoRE is an MoE-based MTL model, this table directly contradicts the headline claim on this dataset. The claim as written is false if the table is correct.
  2. [Section 4.2 and Table 1] The text in Section 4.2 states that with HRNet18, FGMoE achieves the highest SemSeg mIoU (68.12) and PartSeg mIoU (63.03) on PASCAL-Context, but Table 1 lists the FGMoE HRNet18 row as SemSeg 60.52 and PartSeg 59.01, which are lower than MTI-Net's 61.70 and 60.18. The text and table are irreconcilably inconsistent; the claim of state-of-the-art performance in this setting is unsupported as printed.
  3. [Section 4.1 and Table 2] Recomputing Δm from Table 2 using the formula in Section 4.1, with lower-is-better metrics inverted so that higher is better, gives approximately -0.85% for FGMoE ((56.16-56.77)/56.77 + (0.5141-0.5071)/0.5141 + (18.56-19.19)/18.56 + (78.70-78.93)/78.93, averaged over four tasks), not the reported -0.58%. The same computation for MLoRE yields approximately +0.11%, matching the table. This is an arithmetic or reporting error in the paper's primary aggregate metric.
  4. [Tables 4 and 5] Tables 4 and 5 both report experiments on NYUD-v2 with a Swin-T backbone, but they list different single-task baselines: Table 4 gives SemSeg 33.18 and Depth rmse 0.667, while Table 5 gives SemSeg 42.92 and Depth rmse 0.6104. Unless a different training protocol or task set is being used, these baselines should be identical; without clarification, the fine-tuning gains in Table 4 cannot be compared with the full-model ablation in Table 5.
  5. [Section 3.4, Eqs. (10)-(13)] The text after Eq. (13) says that the intra-task expert design "increases the number of non-zero gates to N_it, enabling full activation of all task-specific experts," but Eq. (12) defines a top-k gate that produces exactly K non-zero values. With the default K=3 and N_it=6 used in the experiments, six experts are not fully activated; the method description is internally inconsistent with the routing equations and with the top-k ablation in Table 7.
minor comments (5)
  1. [Throughout] There are several typos and grammatical errors: "Th overall" in Section 3.1, "miminzing" in Section 2.2, "mxier" in Section 4.3, and "cures" for "captures" in Section 4.4. The paper should be carefully proofread.
  2. [Table 3] The method label "Adapte" should be "Adapter". Also, the single-task baseline in Table 3 (SemSeg 67.21, PartSeg 61.93, Sal 62.35, Normal 17.97) differs substantially from the HRNet18 baseline in Table 1; the backbone used for the Table 3 experiments should be stated explicitly.
  3. [Section 4.2] The sentence "we re-implement several leading CNN-based multi-task learning methods on the ViT-L backbone" is ambiguous: it is unclear whether the ViT-L rows for PAD-Net, MTI-Net, ATRC, and InvPT in Table 1 are official results, re-implementations, or evaluations with the official code. The '†' footnote appears only for the Swin-L InvPT row.
  4. [Section 4.4] The qualitative discussion contains an incomplete sentence: "it frequently maintain contextual coherence in complex scenes" lacks a verb form consistent with the subject. The paragraph should be revised.
  5. [Section 4.1] The definition of Δm does not specify how lower-is-better metrics (rmse, mErr) are incorporated. The paper should state explicitly that these metrics are inverted, e.g., using (F_s - F_m)/F_s, so that Δm is consistently higher-is-better.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: FGMoE's claims rest on direct experiments and ablations, not on self-referential derivation or fitted-input prediction.

full rationale

This paper is an empirical architecture paper. The central claim is that FGMoE 'uses fewer parameters and significantly outperforms current MoE-based competitive MTL models on two dense prediction datasets,' and this claim is supported by reported benchmark comparisons, ablations, and parameter counts. None of the load-bearing steps reduces, by the paper's own equations or by self-citation, to its own inputs. The design of the deformable mixer is credited to a prior paper by one of the authors ('The design of deformable mxier (DM) block is inspired by DeMT [39], and we refine its structure'), but this is not used to justify the reported improvements; instead, Table 5 directly ablates the deformable mixer, global MoEs, and task-specific MoEs. The other self-citations are survey or related-work references and are not load-bearing. The evaluation metric Δm is a standard per-task performance drop formula, and the hyper-parameters are stated explicitly; there is no fitted parameter that is later renamed as a prediction. No uniqueness theorem is imported from the authors' prior work, and no ansatz is smuggled in via citation. The paper's internal numerical inconsistencies, such as Table 2 showing FGMoE with worse surface-normal mErr than MLoRE on NYUD-v2 and the §4.2 HRNet18 results disagreeing with Table 1, are serious correctness concerns but they are not circularity: they concern whether the reported evidence is accurate, not whether the derivation is equivalent to its inputs. Under the hard rules, circularity cannot be claimed without exhibiting a specific reduction of a claimed result to an input, and no such reduction exists here. The appropriate finding is no significant circularity, score 0.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

No fundamental physical or conceptual entities are introduced; the new modules are architectural components whose evidence is internal to the paper, so the invented-entities ledger is left empty.

free parameters (5)
  • Loss balancing weights β_t = β_seg=1.0, β_depth=1.0, β_normal=10.0, β_bound=50.0, β_partseg=2.0, β_sal=5.0
    Chosen by hand to balance task losses; changes in these weights would change relative task performance, and no sensitivity analysis is provided.
  • Number of intra-task routing experts = 6
    Ablation in Table 6 shows 8 experts gives higher Δm (6.35 vs 5.17), yet the default uses 6; this choice is not the best in the paper's own ablation.
  • Number of shared experts = 2
    Fixed to 2; no ablation on this number is reported.
  • Top-K routing = 3 (implicit from default config; Table 7 varies K)
    Ablation shows K=4 gives higher Δm (5.60) than the default K=3 (5.17); the default is not the best in the paper's own Table 7.
  • Expert intermediate hidden size = 256
    Chosen without ablation; affects parameter count and capacity.
assumptions (4)
  • domain assumption Multi-task dense prediction tasks share common features that can be learned jointly from a shared frozen backbone.
    The whole decoder-only fine-tuning design assumes a single frozen Swin encoder provides sufficient generic features; not proven, Section 3.1.
  • ad hoc to paper Partitioning expert MLPs along hidden dimensions produces non-overlapping specialized subspaces.
    The central claim of fine-grained specialization rests on this, but no analysis or measurement of expert overlap is provided, Section 3.4.
  • ad hoc to paper The global MoE layer prevents harmful interference while enabling positive transfer.
    No mechanistic evidence is given; the claim is supported only by end-to-end ablations, Section 3.5.
  • standard math Standard transformer and deformable convolution machinery works as expected under the proposed routing.
    Uses off-the-shelf Swin Transformer, MHSA, deformable conv; no proofs needed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multi-Task Dense Prediction Fine-Tuning with Mixture of Fine-Grained Experts." pith.science (2026). https://pith.science/paper/JIG42SMC

@misc{pith2026250719077,
  author       = {Pith},
  title        = {Pith review of: Multi-Task Dense Prediction Fine-Tuning with Mixture of Fine-Grained Experts},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JIG42SMC}},
  note         = {Machine review of arXiv:2507.19077}
}
read the original abstract

Multi-task learning (MTL) for dense prediction has shown promising results but still faces challenges in balancing shared representations with task-specific specialization. In this paper, we introduce a novel Fine-Grained Mixture of Experts (FGMoE) architecture that explores MoE-based MTL models through a combination of three key innovations and fine-tuning. First, we propose intra-task experts that partition along intermediate hidden dimensions of MLPs, enabling finer decomposition of task information while maintaining parameter efficiency. Second, we introduce shared experts that consolidate common information across different contexts of the same task, reducing redundancy, and allowing routing experts to focus on unique aspects. Third, we design a global expert that facilitates adaptive knowledge transfer across tasks based on both input feature and task requirements, promoting beneficial information sharing while preventing harmful interference. In addition, we use the fine-tuning approach to improve parameter efficiency only by training the parameters of the decoder. Extensive experimental results show that the proposed FGMoE uses fewer parameters and significantly outperforms current MoE-based competitive MTL models on two dense prediction datasets (\textit{i.e.,} NYUD-v2, PASCAL-Context) in various metrics.

Figures

Figures reproduced from arXiv: 2507.19077 by the authors.

Figure 1
Figure 1. Overview of our FGMoE model. Given an input image, the Transformer Encoder extracts features that are initially [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. An overview of the task-specific MoEs block. Task-specific MoEs consists of the shared experts, Top-k, intra-task [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Qualitative results on 5-task PASCAL-Context [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

51 extracted references · 43 canonical work pages

  1. [1]

    Ahmed Agiza, Marina Neseem, and Sherief Reda. 2024. MTLoRA: A Low-Rank Adaptation Approach for Efficient Multi-Task Learning. In IEEE/CVF Conference on Computer Vision and Pattern Recognition . 16196–16205. Multi-Task Dense Prediction Fine-Tuning with Mixture of Fine-Grained Experts

    work page 2024

  2. [2]

    Deblina Bhattacharjee, Sabine Süsstrunk, and Mathieu Salzmann. 2023. Vision transformer adapters for generalizable multitask learning. In Proceedings of the IEEE/CVF International Conference on Computer Vision . 19015–19026

    work page 2023

  3. [3]

    David Bruggemann, Menelaos Kanakis, Anton Obukhov, Stamatios Georgoulis, and Luc Van Gool. 2021. Exploring relational context for multi-task dense prediction.. In Proceedings of the IEEE/CVF International Conference on Computer Vision. 15869–15878

    work page 2021

  4. [4]

    Xianjie Chen, Roozbeh Mottaghi, Xiaobai Liu, Sanja Fidler, Raquel Urtasun, and Alan Yuille. 2014. Detect what you can: Detecting and representing objects using holistic models and body parts. In IEEE/CVF Conference on Computer Vision and Pattern Recognition. 1971–1978

    work page 2014

  5. [5]

    Zitian Chen, Yikang Shen, Mingyu Ding, Zhenfang Chen, Hengshuang Zhao, Erik G Learned-Miller, and Chuang Gan. 2023. Mod-squad: Designing mixtures of experts as modular multi-task learners. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 11828–11837

    work page 2023

  6. [6]

    Damai Dai, Chengqi Deng, Chenggang Zhao, RX Xu, Huazuo Gao, Deli Chen, Jiashi Li, Wangding Zeng, Xingkai Yu, Yu Wu, et al. 2024. Deepseekmoe: Towards ultimate expert specialization in mixture-of-experts language models. arXiv preprint arXiv:2401.06066 (2024)

    arXiv 2024

  7. [7]

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xi- aohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, and Neil Houlsby. 2021. An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale. InInternational Conference on Learning Representations

    2021

  8. [8]

    Nan Du, Yanping Huang, Andrew M Dai, Simon Tong, Dmitry Lepikhin, Yuanzhong Xu, Maxim Krikun, Yanqi Zhou, Adams Wei Yu, Orhan Firat, et al

Show all 51 references
  1. [9]

    William Fedus, Barret Zoph, and Noam Shazeer. 2022. Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity. Journal of Machine Learning Research 23, 120 (2022), 1–39

    2022

  2. [10]

    Junxian He, Chunting Zhou, Xuezhe Ma, Taylor Berg-Kirkpatrick, and Graham Neubig. 2022. Towards a unified view of parameter-efficient transfer learning. In The Tenth International Conference on Learning Representations

    2022

  3. [11]

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2022. Lora: Low-rank adaptation of large language models. In The Tenth International Conference on Learning Representa- tions

    2022

  4. [12]

    Huimin Huang, Yawen Huang, Lanfen Lin, Ruofeng Tong, Yen-Wei Chen, Hao Zheng, Yuexiang Li, and Yefeng Zheng. 2024. Going beyond multi-task dense prediction with synergy embedding models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 28181–28190

    2024

  5. [13]

    Robert A Jacobs and Michael I Jordan. 1993. Learning piecewise control strategies in a modular neural network architecture. IEEE Transactions on Systems, Man, and Cybernetics 23, 2 (1993), 337–345

    1993

  6. [14]

    Robert A Jacobs, Michael I Jordan, Steven J Nowlan, and Geoffrey E Hinton. 1991. Adaptive mixtures of local experts. Neural computation 3, 1 (1991), 79–87

    1991

  7. [15]

    Menglin Jia, Luming Tang, Bor-Chun Chen, Claire Cardie, Serge Belongie, Bharath Hariharan, and Ser-Nam Lim. 2022. Visual prompt tuning. In Euro- pean Conference on Computer Vision . 709–727

    2022

  8. [16]

    Fengze Jiang, Shuling Wang, and Xiaojin Gong. 2024. Task-Conditional Adapter for Multi-Task Dense Prediction. In Proceedings of the 32nd ACM International Conference on Multimedia. 2059–2068

    2024

  9. [17]

    Rabeeh Karimi Mahabadi, James Henderson, and Sebastian Ruder. 2021. Com- pacter: Efficient low-rank hypercomplex adapter layers. In Advances in Neural Information Processing Systems, Vol. 34. 1022–1035

    2021

  10. [18]

    Iasonas Kokkinos. 2017. Ubernet: Training a universal convolutional neural network for low-, mid-, and high-level vision using diverse datasets and limited memory. In IEEE/CVF Conference on Computer Vision and Pattern Recognition . 6129–6138

    2017

  11. [19]

    Dmitry Lepikhin, HyoukJoong Lee, Yuanzhong Xu, Dehao Chen, Orhan Firat, Yanping Huang, Maxim Krikun, Noam Shazeer, and Zhifeng Chen. 2021. GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding. In 9th International Conference on Learning Representations

    2021

  12. [20]

    Hanxue liang, Zhiwen Fan, Rishov Sarkar, Ziyu Jiang, Tianlong Chen, Kai Zou, Yu Cheng, Cong Hao, and Zhangyang Wang. 2022. M3ViT: Mixture-of-Experts Vision Transformer for Efficient Multi-task Learning with Model-Accelerator Co-design. In Advances in Neural Information Process...

    2022

  13. [21]

    Xiwen Liang, Yangxin Wu, Jianhua Han, Hang Xu, Chunjing Xu, and Xiaodan Liang. 2022. Effective adaptation in multi-task co-training for unified autonomous driving. Advances in Neural Information Processing Systems 35 (2022), 19645– 19658

    2022

  14. [22]

    Hanxiao Liu, Zihang Dai, David So, and Quoc V Le. 2021. Pay Attention to MLPs. In Advances in neural information processing systems , Vol. 34. 9204–9215

    2021

  15. [23]

    Yen-Cheng Liu, Chih-Yao Ma, Junjiao Tian, Zijian He, and Zsolt Kira. 2022. Polyhistor: Parameter-Efficient Multi-Task Adaptation for Dense Vision Tasks. In Advances in Neural Information Processing Systems

    2022

  16. [24]

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. 2021. Swin Transformer: Hierarchical Vision Transformer using Shifted Windows. In Proceedings of the IEEE/CVF International Conference on Computer Vision. 10012–10022

    2021

  17. [25]

    Yuxiang Lu, Shalayiding Sirejiding, Yue Ding, Chunlin Wang, and Hongtao Lu. 2024. Prompt guided transformer for multi-task dense prediction. IEEE Transactions on Multimedia 26 (2024), 6375–6385

    2024

  18. [26]

    Rabeeh Karimi Mahabadi, Sebastian Ruder, Mostafa Dehghani, and James Hen- derson. 2021. Parameter-efficient Multi-task Fine-tuning for Transformers via Shared Hypernetworks. In ACL/IJCNLP. 565–576

    2021

  19. [27]

    Krishna Sri Ipsit Mantri, Carola-Bibiane Schönlieb, Bruno Ribeiro, Chaim Baskin, and Moshe Eliasof. 2025. DiTASK: Multi-Task Fine-Tuning with Diffeomor- phic Transformations. In IEEE/CVF Conference on Computer Vision and Pattern Recognition

    2025

  20. [28]

    Yuzhang Shang, Dan Xu, Gaowen Liu, Ramana Rao Kompella, and Yan Yan

  21. [29]

    Nathan Silberman, Derek Hoiem, Pushmeet Kohli, and Rob Fergus. 2012. Indoor segmentation and support inference from rgbd images. In European Conference on Computer Vision. 746–760

    2012

  22. [30]

    Ke Sun, Bin Xiao, Dong Liu, and Jingdong Wang. 2019. Deep High-Resolution Representation Learning for Human Pose Estimation. In IEEE/CVF Conference on Computer Vision and Pattern Recognition . 5693–5703

    2019

  23. [31]

    Yi-Lin Sung, Jaemin Cho, and Mohit Bansal. 2022. Vl-adapter: Parameter-efficient transfer learning for vision-and-language tasks. In IEEE/CVF Conference on Com- puter Vision and Pattern Recognition . 5227–5237

    2022

  24. [32]

    Anke Tang, Li Shen, Yong Luo, Shuai Xie, Han Hu, Lefei Zhang, Bo Du, and Dacheng Tao. 2024. Smile: Zero-shot sparse mixture of low-rank experts con- struction from pre-trained foundation models. arXiv preprint arXiv:2408.10174 (2024)

    2024 arXiv

  25. [33]

    Anke Tang, Li Shen, Yong Luo, Nan Yin, Lefei Zhang, and Dacheng Tao. 2024. Merging Multi-Task Models via Weight-Ensembling Mixture of Experts. In Inter- national Conference on Machine Learning . PMLR, 47778–47799

    2024

  26. [34]

    Vandenhende, S

    S. Vandenhende, S. Georgoulis, W. Van Gansbeke, M. Proesmans, D. Dai, and L. Van Gool. 2022. Multi-Task Learning for Dense Prediction Tasks: A Survey. IEEE TPAMI 44, 7 (2022), 3614–3633

    2022

  27. [35]

    Simon Vandenhende, Stamatios Georgoulis, Luc Van Gool, and Luc Van Gool

  28. [36]

    Yi Xin, Junlong Du, Qiang Wang, Zhiwen Lin, and Ke Yan. 2024. Vmt-adapter: Parameter-efficient transfer learning for multi-task dense scene understanding. In Proceedings of the AAAI Conference on Artificial Intelligence , Vol. 38. 16085– 16093

    2024

  29. [37]

    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 IEEE/CVF Conference on Computer Vision and Pattern Recognition. 675–684

    2018

  30. [38]

    Yangyang Xu, Xiangtai Li, Haobo Yuan, Yibo Yang, and Lefei Zhang. 2024. Multi- Task Learning with Multi-query Transformer for Dense Prediction. IEEE Trans- actions on Circuits and Systems for Video Technology 34, 2 (2024), 1228–1240

    2024

  31. [39]

    Yangyang Xu, Yibo Yang, and Lefei Zhang. 2023. DeMT: Deformable Mixer Transformer for Multi-Task Learning of Dense Prediction. In Proceedings of the AAAI Conference on Artificial Intelligence . 3072–3080

    2023

  32. [40]

    Yangyang Xu, Yibo Yang, and Lefei Zhang. 2023. Multi-task learning with knowl- edge distillation for dense prediction. InProceedings of the IEEE/CVF International Conference on Computer Vision . 21550–21559

    2023

  33. [41]

    Siwei Yang, Hanrong Ye, and Dan Xu. 2023. Contrastive multi-task dense pre- diction. In Proceedings of the AAAI Conference on Artificial Intelligence , Vol. 37. 3190–3197

    2023

  34. [42]

    Yuqi Yang, Peng-Tao Jiang, Qibin Hou, Hao Zhang, Jinwei Chen, and Bo Li. 2024. Multi-task dense prediction via mixture of low-rank experts. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition . 27927–27937

    2024

  35. [43]

    Yuqi Yang, Peng-Tao Jiang, Qibin Hou, Hao Zhang, Jinwei Chen, and Bo Li. 2025. Multi-Task Dense Predictions via Unleashing the Power of Diffusion. In The Thirteenth International Conference on Learning Representations

    2025

  36. [44]

    Hanrong Ye and Dan Xu. 2022. Inverted Pyramid Multi-task Transformer for Dense Scene Understanding. In European Conference on Computer Vision . 514– 530

    2022

  37. [45]

    Hanrong Ye and Dan Xu. 2023. Taskexpert: Dynamically assembling multi-task representations with memorial mixture-of-experts. InProceedings of the IEEE/CVF International Conference on Computer Vision . 21828–21837

    2023

  38. [46]

    Hanrong Ye and Dan Xu. 2023. TaskPrompter: Spatial-Channel Multi-Task Prompting for Dense Scene Understanding. In The Eleventh International Confer- ence on Learning Representations

    2023

  39. [47]

    Hanrong Ye and Dan Xu. 2024. Diffusionmtl: Learning multi-task denoising diffusion model from partially annotated data. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 27960–27969. Yangyang Xu et al

    2024

  40. [48]

    Elad Ben Zaken, Yoav Goldberg, and Shauli Ravfogel. 2022. BitFit: Sim- ple Parameter-efficient Fine-tuning for Transformer-based Masked Language- models. In ACL. 1–9

    2022

  41. [2020]

    In European Conference on Computer Vision

    Mti-net: Multi-scale task interaction networks for multi-task learning. In European Conference on Computer Vision . 527–543

  42. [2022]

    In International conference on machine learning

    Glam: Efficient scaling of language models with mixture-of-experts. In International conference on machine learning . PMLR, 5547–5569

  43. [2024]

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

    Efficient multitask dense predictor via binarization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 15899–15908

Pith tools

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