Pith. sign in

REVIEW 4 major objections 5 minor 55 references

FEAST, a reinforcement-learning-guided segmentation decoder, sets a new state of the art on a novel 24k-image dish benchmark, beating Mask2Former by 3.21 mIoU.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 03:40 UTC pith:NKRKDHPU

load-bearing objection The dish dataset is a real contribution; the RL-decoder story is mostly MoE, and the headline +3.21% needs an honest accounting of the pretrained reference policy before it can be trusted. the 4 major comments →

arxiv 2607.23070 v1 pith:NKRKDHPU submitted 2026-07-25 cs.CV

DishSeg24k: A Large-Scale Benchmark for Food Segmentation with Stochastic Expert Decoding

Yilin Wang , Haochen Shi , Guanyu Chen , Weiqing Min , Jinkai Zheng , Chenggang Yan , Shuqiang Jiang This is my paper
classification cs.CV
keywords food segmentationbenchmarkinstance segmentationlong-tail distributionmixture-of-expertsMarkov decision processreinforcement learningDishSeg24k
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper introduces DishSeg24k, a large-scale benchmark for dish-level food segmentation with 24,096 real-world dining images, 112,281 annotated instances across 278 fine-grained categories, built around dense inter-dish overlap, fine-grained class similarity, and a long-tail distribution. The authors argue that existing benchmarks oversimplify real dining scenes and that query-based segmentation decoders are too deterministic. They propose FEAST, which models query refinement as a Markov Decision Process where each decoder layer applies a continuous 'semantic shift' sampled from a policy, routed through a mixture of shared and specialized experts, and trained with two decoupled critics (task and structure rewards). On DishSeg24k, FEAST with ResNet-50 reaches 53.43% mIoU, outperforming the strongest baseline Mask2Former by +3.21% mIoU, and it also improves FoodSeg103 by +1.47% mIoU. If the reported numbers are accurate, FEAST is the best-performing food segmentation method on both benchmarks, and DishSeg24k becomes the most demanding public testbed for dish-level segmentation.

Core claim

The central claim is that modeling query-based mask decoding as a sequential stochastic decision process—rather than deterministic layer-by-layer refinement—improves segmentation in densely overlapping, long-tailed food scenes. FEAST treats each decoder layer as an action that shifts object queries in embedding space (q_{k+1}=q_k+a_k), with the action sampled from a Gaussian distribution produced by an MoE policy network. Two independent reward signals guide the training: a task reward measuring incremental mask and classification improvement, and a structure reward encouraging balanced expert routing. The authors show that the MoE decoder alone gives the largest gain (+3.03% mIoU over the M

What carries the argument

The key mechanism is the stochastic MoE decoder: object queries are updated via Gaussian-sampled semantic shifts in a residual transition, routed through one shared expert plus E_r routed experts with a stop-gradient router, and trained with dual critics (task and structure) via policy gradient. A KL penalty anchors the learned policy to a frozen pretrained Mask2Former decoder to prevent catastrophic deviation, while the structure reward (routing entropy minus a load-balancing KL) prevents expert collapse under the long-tail distribution.

Load-bearing premise

All method comparisons on DishSeg24k inherit the annotation rule that overlapping ingredients are assigned to the dominant dish, which presumes a single well-defined label for every pixel in dense overlap; no inter-annotator agreement is reported to verify that this rule is applied consistently.

What would settle it

Take a random subset of DishSeg24k test images (e.g., 500) and have a second set of annotators re-label them under the same protocol; compute pixel-level agreement. If agreement is significantly lower in overlapping regions than in non-overlapping regions, or if the mIoU ranking of methods changes when evaluated on a high-agreement subset, then the reported differences between methods may be dominated by annotation noise rather than segmentation quality.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If the reported gains are accurate, FEAST establishes a new state of the art for food segmentation, beating Mask2Former by +3.21 mIoU on DishSeg24k and +1.47 mIoU on FoodSeg103.
  • DishSeg24k becomes the largest dish-level benchmark to date (24,096 images, 112,281 instances, 278 categories, 4.66 instances per image), offering a more demanding testbed for food segmentation and downstream tasks like volume estimation and dietary assessment.
  • The RL-guided MoE decoding framework is general: it can be applied to other segmentation tasks that exhibit dense overlap and long-tail distributions, not just food.
  • The decoupled dual-reward design—separating mask-quality improvement from routing-balance—provides a recipe for stabilizing MoE training under imbalanced data and may generalize to other long-tail recognition problems.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The benchmark's annotation rule for overlapping ingredients ('unified masks based on dominant dish identity') presumes a single correct label for every pixel; without inter-annotator agreement statistics, the reported mIoU differences between methods could partly reflect label noise rather than pure segmentation ability.
  • At inference, FEAST sets the action to its mean (a_k = mu_k), making the decoder deterministic; the stochastic exploration benefits are realized only during training. An extension would test whether sampling multiple actions at test time and ensembling the resulting masks improves boundary accuracy in overlap regions.
  • A direct diagnostic experiment would be to re-annotate a random subset of DishSeg24k test images by a second set of annotators and compute pixel-level agreement; if agreement is low in overlapping regions, the method ranking may shift when evaluated on a high-agreement subset.
  • The dataset's additionally mentioned ingredient-level annotations could enable a transfer-learning test: pretraining on dish-level masks and then evaluating on ingredient-level benchmarks (or vice versa) would quantify how well the dish-level representation supports finer-grained understanding.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper introduces DishSeg24k, a large-scale dish-level food segmentation benchmark with 24,096 images, 112,281 instances, and 278 categories, plus a proposed method FEAST that formulates query-based decoding as an MDP with an RL-guided Mixture-of-Experts decoder. The authors report that FEAST outperforms Mask2Former by +3.21% mIoU on DishSeg24k with a ResNet-50 backbone and generalizes to FoodSeg103 (+1.47% mIoU). The central claim is that the stochastic expert-decoding mechanism is responsible for these gains, and that the new benchmark captures dense overlap, fine-grained similarity, and long-tail distributions.

Significance. If the benchmark is of high annotation quality and the method's gains are real, DishSeg24k would be a valuable resource for food segmentation research, and the RL-guided MoE decoder could be a useful architectural alternative. The paper includes a large-scale dataset with detailed statistics, extensive comparisons, code/model release planned, and a fairly complete ablation study. However, the significance is tempered by two unresolved issues: the marginal contribution of the RL components themselves (Table 4: R5 vs R0 is only +0.18% mIoU) and the undisclosed pretraining of the reference policy, which could confound the headline SOTA comparison.

major comments (4)
  1. [Sec. 5.1, Eq. (6)] The reference policy π_ref is described only as 'the frozen pretrained Mask2Former decoder,' but the pretraining data is never stated. If π_ref was pretrained on DishSeg24k (or another food segmentation dataset), then FEAST receives a pretraining advantage that the Mask2Former baseline in Table 2 does not receive, since the baselines are 'trained from scratch' under the stated protocol. This would directly confound the +3.21% mIoU headline gain and the +3.03% MoE ablation gain in Table 3 (B vs A). Please disclose the pretraining data for π_ref and rerun the Mask2Former baseline initialized from the same checkpoint, or otherwise demonstrate that the comparison is fair.
  2. [Sec. 4.4, Eq. (4), Table 4] The task reward in Eq. (4) is computed from the same ground-truth masks and the same final-layer bipartite matching used by the segmentation loss L_seg, making it closer to per-layer deep supervision than to an independent RL signal. The claimed central mechanism—RL-guided stochastic decoding—is weakly supported: Table 4 shows that each individual RL component (R1, R2, R3, R4) degrades performance below the MoE-only reference, and the full RL stack (R5) adds only +0.18% mIoU over MoE-only (R0). The paper should either provide direct evidence of the RL mechanism's effect (e.g., reward curves, query trajectory diversity, or a comparison against a non-RL stochastic MoE baseline) or temper the claim that RL decoding is the driving contribution.
  3. [Sec. 3.2, Annotation Workflow] The annotation rule that 'overlapping ingredients are assigned unified masks based on dominant dish identity' presumes that a dominant-dish label is well-defined for every pixel in dense overlap. No inter-annotator agreement or label-noise study is reported for this critical rule. If the rule is applied inconsistently, the small metric differences between methods (including the +0.18% mIoU from the RL components) could be within annotation noise. Please report a label-quality study on a subset (e.g., pixel-level agreement on overlapping regions) to justify the benchmark's use for fine-grained method comparison.
  4. [Sec. 5.4, Table 4] Table 4's mDice and mAcc columns appear to be swapped relative to Table 3 (Row B: 53.25 mIoU, 60.48 mDice, 60.68 mAcc, whereas Table 4 R0 shows 60.68 mDice and 60.48 mAcc). Please verify the reported numbers; if the columns are swapped, the ablation conclusions about mDice/mAcc need to be corrected.
minor comments (5)
  1. [Abstract / Sec. 5.1] FoodSeg103 is described as having 104 categories in Sec. 5.5 but 103 in the abstract and elsewhere; please standardize the count.
  2. [Fig. 4 caption] Typo: 'Doupled rewards' should be 'Decoupled rewards'.
  3. [Table 5] CCNet appears twice with different reference numbers ([15] and [37]); the original CCNet is [15], while [37] is the FoodSeg103 benchmark paper. Please disambiguate.
  4. [Table 1] The 'UECFoodPixCompomplete' entry has a typo; should be 'UECFoodPixComplete'.
  5. [Sec. 4.2] The MDP is defined as a tuple (S,A,P,R,γ), but no transition probability distribution is specified for stochastic policies; since the transition is deterministic, this should be stated explicitly.

Circularity Check

0 steps flagged

No significant circularity: FEAST's performance claims are empirical benchmark results; the RL reward is a supervised training signal, not a fitted prediction, and the self-cited ZSFood source is reused only for raw images.

full rationale

The paper's central claims are empirical: FEAST achieves higher mIoU on a new benchmark and on the external FoodSeg103 benchmark. There is no derivation chain in which a predicted quantity is defined in terms of the fitted input. The task reward in Eq. (4) uses the same ground-truth masks and bipartite matching as the segmentation loss, but this is standard supervised training signal (deep supervision), not a separate prediction claimed to be derived from first principles. The method does not fit a parameter to a subset and then 'predict' a closely related quantity. The only self-citation of note is [45] ZSFood, whose raw images are reused for DishSeg24k after discarding the original bounding-box annotations and re-annotating pixels; this is a disclosed data source, not an unverified load-bearing theoretical premise. No uniqueness theorem is imported from the authors' prior work, and no ansatz is smuggled via self-citation. The undisclosed pretraining of the reference policy π_ref is a legitimate experimental-transparency and fairness concern, but it is not circularity: the paper does not claim to derive the reference policy, and the reported comparisons could be unfair without making the result definitionally equivalent to its inputs. The benchmark evaluation is external to the method, and the FoodSeg103 results provide an independent, externally falsifiable check. We therefore find no circular step requiring a score above zero.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 0 invented entities

The central claims rest on annotation-quality assumptions and unreported RL hyperparameters; no new physical entities are introduced. The benchmark's validation relies on a subjective taxonomy and a dominant-dish labeling rule that is not independently checked.

free parameters (5)
  • Reward weights λ_dice, λ_cls, λ_div, λ_bal
    Introduced in Eqs. (4) and (5) to combine task and structure rewards; numerical values are not reported, and no sensitivity analysis is given, so the RL objective is under-specified.
  • KL penalty coefficient β = 0.01
    Used in Eq. (6) to anchor the learned policy to a frozen Mask2Former decoder; chosen by hand with no ablation.
  • GAE λ = 0.95
    Sec. 4.4 fixes λ_GAE=0.95 without sensitivity analysis.
  • Routing temperature τ schedule = 1.0 → 0.01 annealed linearly
    Eq. (1); the schedule details and its effect on routing are not ablated.
  • Number of routed experts E_r = 4
    Selected by the Fig. 6 ablation (2/4/8 experts); this is central to the MoE gain.
axioms (5)
  • standard math Mask2Former's masked cross-attention, self-attention, bipartite matching, and per-query losses are a working base (Sec. 4.1, 4.4).
    The method is built on the Mask2Former pipeline and assumes its query-based segmentation machinery is correct.
  • domain assumption The contextual MDP Markov property holds with pixel features as static context and query state s_k (Sec. 4.2).
    The state omits the policy's stochastic history; this is a modeling choice, not proven.
  • ad hoc to paper A single forward pass is a complete K-step episode and one parameter update per batch yields a valid on-policy policy gradient (Sec. 4.5, 5.1).
    REINFORCE/GAE with one episode per update can be high-variance; stability is assumed and not analyzed.
  • domain assumption Overlapping dishes have a well-defined 'dominant dish identity' mask (Sec. 3.2, Annotation Workflow).
    No inter-annotator agreement is reported; label noise would directly affect all benchmark numbers.
  • domain assumption Images from ZSFood plus 3,493 collected web images represent real-world dining (Sec. 3.1).
    No sampling or demographic description beyond Chinese/Western restaurant scenes is provided.

pith-pipeline@v1.3.0-alltime-deepseek · 13898 in / 17110 out tokens · 152619 ms · 2026-08-01T03:40:55.595730+00:00 · methodology

0 comments
read the original abstract

Food segmentation is essential for applications such as intelligent catering, dietary assessment, and recommendation. However, existing benchmarks fail to capture the complexity of real-world dining scenes. The challenges of dense inter-dish overlap, fine-grained class similarity, and extreme long-tail class distributions exceed the fidelity of current datasets. To fill this gap, we introduce \textbf{DishSeg24k}, a large-scale dish-level segmentation benchmark with 24,096 images, 112,281 instances, and 278 fine-grained categories in real-world dining environments. Based on DishSeg24k, we further propose \textbf{Food Expert-Adaptive Segmentation Transformers (FEAST)} to address these challenges. FEAST models query-based decoding as a Markov Decision Process (MDP), where each decoder layer update is treated as a sequential decision step that explores uncertainty along dish boundaries. We further redesign the decoder with a reinforcement learning (RL)-guided Mixture-of-Experts (MoE) module, in which a dual-critic decoupled optimization scheme separates task-oriented query refinement from structure-aware expert routing. This design promotes expert specialization and prevents expert collapse under long-tail category distributions. Finally, extensive experiments on DishSeg24k demonstrate the state-of-the-art performance of FEAST, which outperforms previous methods by {+3.21\%} mIoU, {+3.68\%} mDice, and {+4.00\%} mAcc, respectively. We further validate the effectiveness of FEAST on FoodSeg103. The dataset and code will be publicly released.

Figures

Figures reproduced from arXiv: 2607.23070 by Chenggang Yan, Guanyu Chen, Haochen Shi, Jinkai Zheng, Shuqiang Jiang, Weiqing Min, Yilin Wang.

Figure 1
Figure 1. Figure 1: Cross-dataset qualitative comparison. For each [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Representative annotated samples from DishSeg24k across diverse serving styles, which range from single plates to [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Statistical analysis of DishSeg24k. (a) Category distribution exhibits a pronounced long-tail pattern. The upper-right [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Overview of FEAST with four components. (a) Overall pipeline. (b) SRMoE within stochastic decoder layer. (c) MDP [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: (a) Qualitative segmentation comparison between methods. (b) Activation maps of MoE experts. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Effect of the number of MoE routing experts (%). [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

55 extracted references · 6 linked inside Pith

  1. [1]

    G. M. Abhilash, M. Eshwar, S. Achuth, and M. N. Harshitha. 2025. NutriVision: YOLO-SAM-Driven Dietary App for Personalized Food Recommendation and Meal Tracking. In2025 International Conference on Innovative Trends in Informa- tion Technology

    2025

  2. [2]

    Jatin Alla, Yashas Samaga, Ashwin Vaswani, et al. 2026. FoodLens: Fine-Grained and Multi-label Classification of Indian Food Images. InImage Processing and Vision Engineering. Springer Nature Switzerland, Cham, 262–284

    2026

  3. [3]

    Ahmad AlMughrabi, Umair Haroon, Ricardo Marques, and Petia Radeva. 2025. VolTex: Food Volume Estimation using Text-Guided Segmentation and Neu- ral Surface Reconstruction. InProceedings of the Computer Vision and Pattern Recognition Conference. 450–457

    2025

  4. [4]

    Marc Bolaños, Aina Ferrà, and Petia Radeva. 2017. Food Ingredients Recognition through Multi-Label Learning. InNew Trends in Image Analysis and Processing

    2017

  5. [5]

    Jingjing Chen, Bin Zhu, Chong-Wah Ngo, Tat-Seng Chua, and Yu-Gang Jiang

  6. [6]

    Liang-Chieh Chen, Yukun Zhu, George Papandreou, Florian Schroff, and Hartwig Adam. 2018. Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation. InEuropean Conference on Computer Vision. 801–818

    2018

  7. [7]

    Schwing, Alexander Kirillov, and Rohit Girdhar

    Bowen Cheng, Ishan Misra, Alexander G. Schwing, Alexander Kirillov, and Rohit Girdhar. 2022. Masked-Attention Mask Transformer for Universal Image Seg- mentation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 1290–1299

    2022

  8. [8]

    Bowen Cheng, Alexander Schwing, and Alexander Kirillov. 2021. Per-Pixel Classification Is Not All You Need for Semantic Segmentation. InAdvances in Neural Information Processing Systems, Vol. 34. 17864–17875

    2021

  9. [9]

    Zewen Chi, Li Dong, Shaohan Huang, Damai Mao, Heyan Wang, and Furu Wei

  10. [10]

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. 2009. Ima- geNet: A Large-Scale Hierarchical Image Database. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Ieee, 248–255

    2009

  11. [11]

    Charles NC Freitas, Filipe R Cordeiro, and Valmir Macario. 2020. MyFood: A Food Segmentation and Classification System to Aid Nutritional Monitoring. In 2020 33rd SIBGRAPI Conference on Graphics, Patterns and Images. IEEE, 234–239

    2020

  12. [12]

    Yunxiang Fu, Meng Lou, and Yizhou Yu. 2025. SegMAN: Omni-scale Context Mod- eling with State Space Models and Local Attention for Semantic Segmentation. arXiv:2412.11890 [cs.CV]

    Pith/arXiv arXiv 2025

  13. [13]

    Assaf Hallak, Dotan Di Castro, and Shie Mannor. 2015. Contextual Markov Decision Processes.arXiv preprint arXiv:1502.02259(2015)

    Pith/arXiv arXiv 2015

  14. [14]

    Kaiming He, Xiangyu Zhan, Shaoqing Ren, and Jian Sun. 2016. Deep Residual Learning for Image Recognition. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 770–778

    2016

  15. [15]

    Zilong Huang, Xinggang Wang, Lichao Huang, Chang Huang, Yunchao Wei, and Wenyu Liu. 2019. CCNet: Criss-Cross Attention for Semantic Segmentation. In Proceedings of the IEEE/CVF International Conference on Computer Vision. 603–612

    2019

  16. [16]

    Jitesh Jain, Jiachen Li, Mang Tik Chiu, Ali Hassani, Nikita Orlov, and Humphrey Shi. 2023. OneFormer: One Transformer to Rule Universal Image Segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2989–2998

    2023

  17. [17]

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alexander C Berg, Wan-Yen Lo, et al

  18. [18]

    Shanzhen Lan, Chengjuan Wan, Lan Chen, Mingxue Jin, and Shaode Yu. 2022. Deep Learning-Based Recognition of Chinese Dishes in a Waiterless Restaurant. In2022 16th IEEE International Conference on Signal Processing, Vol. 1

    2022

  19. [19]

    Xing Lan, Jiayi Lyu, Hanyu Jiang, Kun Dong, Zehai Niu, Yi Zhang, and Jian Xue

  20. [20]

    Tsung-Yi Lin, Piotr Dollár, Ross Girshick, Kaiming He, Bharath Hariharan, and Serge Belongie. 2017. Feature Pyramid Networks for Object Detection. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 936–944

    2017

  21. [21]

    Xinda Liu, Yaohui Zhu, Linhu Liu, Jiang Tian, and Lili Wang. 2023. Feature- Suppressed Contrast for Self-Supervised Food Pre-training. InProceedings of the 31st ACM International Conference on Multimedia. 4359–4367

    2023

  22. [22]

    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. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). 9992–10002. doi:10.1109/ICCV48922.2021.00986

    arXiv 2021

  23. [23]

    Lujia Lv, Di Wu, Yangyi Xia, Jia Wu, Xiaojing Liu, and Yi He. 2024. Differential Evolution Integrated Hybrid Deep Learning Model for Object Detection in Pre- made Dishes. arXiv:2412.20370 [cs.CV]

    Pith/arXiv arXiv 2024

  24. [24]

    Austin Meyers, Nick Johnston, Vivek Rathod, Anoop Korattikara, Alex Gorban, Nathan Silberman, Sergio Guadarrama, George Papandreou, Jonathan Huang, Conference’17, July 2017, Washington, DC, USA Yilin Wang, Haochen Shi, Guanyu Chen, Weiqing Min, Jinkai Zheng, Chenggang Yan, and Shuqiang Jiang and Kevin P. Murphy. 2015. Im2Calories: Towards an Automated Mob...

    2017

  25. [25]

    Weiqing Min, Zhiling Wang, Yuxin Liu, Mengjiang Luo, Liping Kang, Xiaoming Wei, Xiaodan Wei, and Shuqiang Jiang. 2023. Large Scale Visual Food Recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence45, 8 (2023), 9932– 9949

    2023

  26. [26]

    Sharada Prasanna Mohanty, Gaurav Singhal, Eric Antoine Scuccimarra, Djilani Kebaili, Harris Héritier, Victor Boulanger, and Marcel Salathé. 2022. The Food Recognition Benchmark: Using Deep Learning to Recognize Food in Images. Frontiers in Nutrition9 (2022), 875143

    2022

  27. [27]

    Huu-Thanh Nguyen, Yu Cao, Chong-Wah Ngo, and Wing-Kwong Chan. 2024. FoodMask: Real-time Food Instance Counting,Segmentation and Recognition. Pattern Recognition146 (2024), 110017

    2024

  28. [28]

    Huu Thanh Nguyen, Chong-Wah Ngo, and Wing-Kwong Chan. 2022. SibNet: Food Instance Counting and Segmentation.Pattern Recognition124 (2022), 108470

    2022

  29. [29]

    Zhengyang Ni, Xingang Chen, Yunhai Zhai, Yuhong Tang, and Yuchao Wang

  30. [30]

    Kaimu Okamoto and Keiji Yanai. 2021. UEC-FoodPix Complete: A Large-Scale Food Image Segmentation Dataset. InPattern Recognition. ICPR International Workshops and Challenges. Springer-Verlag, Berlin, Heidelberg

    2021

  31. [31]

    Deepanshu Pandey, Purva Parmar, Gauri Toshniwal, Mansi Goel, Vishesh Agrawal, Shivangi Dhiman, Lavanya Gupta, and Ganesh Bagler. 2022. Ob- ject Detection in Indian Food Platters using Transfer Learning with YOLOv4. arXiv:2205.04841 [cs.CV] https://arxiv.org/abs/2205.04841

    Pith/arXiv arXiv 2022

  32. [32]

    John Schulman, Philipp Moritz, Sergey Levine, Michael Jordan, and Pieter Abbeel

  33. [33]

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov

  34. [34]

    Jong-Hwan Shim, Hyunwoo Yu, Kichan Kong, and Sung Ju Kang. 2023. Feed- Former: Revisiting Transformer Decoder for Efficient Semantic Segmentation. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 37. 2263–2271

    2023

  35. [35]

    Wei Wang. 2023. Advanced Auto Labeling Solution with Added Features. https: //github.com/CVHub520/X-AnyLabeling

    2023

  36. [36]

    Yunan Wang, Jing-jing Chen, Chong-Wah Ngo, Tat-Seng Chua, Wanli Zuo, and Zhaoyan Ming. 2019. Mixed Dish Recognition through Multi-Label Learning. In Proceedings of the 11th Workshop on Multimedia for Cooking and Eating Activities. Association for Computing Machinery, New York, NY, USA

    2019

  37. [37]

    Xiongwei Wu, Xin Fu, Ying Liu, Ee-Peng Lim, Steven C. H. Hoi, and Qianru Sun

  38. [38]

    Xun Wu, Shaohan Huang, Wenhui Wang, Shuming Ma, Li Dong, and Furu Wei

  39. [39]

    Xiongwei Wu, Sicheng Yu, Ee-Peng Lim, and Chong-Wah Ngo. 2024. OVFood- Seg: Elevating Open-Vocabulary Food Image Segmentation via Image-Informed Textual Representation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 4144–4153

    2024

  40. [40]

    Yuxin Wu, Alexander Kirillov, Francisco Massa, Wan-Yen Lo, and Ross Girshick

  41. [41]

    Zhiyong Xiao, Yang Li, and Zhaohong Deng. 2025. Food image segmentation based on deep and shallow dual-branch network.Multimedia Systems31, 2 (Jan. 2025), 85

    2025

  42. [42]

    Alvarez, and Ping Luo

    Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, and Ping Luo. 2021. SegFormer: Simple and Efficient Design for Semantic Segmenta- tion with Transformers. InAdvances in Neural Information Processing Systems, Vol. 34. 12077–12090

    2021

  43. [43]

    Haotian Yan, Ming Wu, and Chuang Zhang. 2024. Multi-scale Representations by Varying Window Attention for Semantic Segmentation. InThe Twelfth Inter- national Conference on Learning Representations

    2024

  44. [44]

    Hyunwoo Yu, Yujin Cho, Beomjun Kang, Seonghyeon Moon, Kichan Kong, and Sung Ju Kang. 2024. Embedding-Free Transformer with Inference Spatial Re- duction for Efficient Semantic Segmentation. InProceedings of the European Conference on Computer Vision. Springer Nature Switzerland, 92–110

    2024

  45. [45]

    InProceedings of the ACM International Conference on Multimedia

    A Large-Scale Benchmark for Food Image Segmentation. InProceedings of the ACM International Conference on Multimedia. 506–515

  46. [47]

    InAdvances in Neural Information Processing Systems

    Multi-head mixture-of-experts. InAdvances in Neural Information Processing Systems. 14205–14215

  47. [55]

    Pengfei Zhou, Weiqing Min, Yang Zhang, Jiajun Song, Ying Jin, and Shuqiang Jiang. 2023. SeeDS: Semantic Separable Diffusion Synthesizer for Zero-shot Food Detection. InProceedings of the 31st ACM International Conference on Multimedia. 8157–8166

    2023

  48. [2015]

    High-Dimensional Continuous Control Using Generalized Advantage Estimation.arXiv preprint arXiv:1506.02438(2015)

    Pith/arXiv arXiv 2015

  49. [2017]

    Proximal Policy Optimization Algorithms.arXiv preprint arXiv:1707.06347 (2017)

    Pith/arXiv arXiv 2017

  50. [2019]

    https://github.com/facebookresearch/detectron2

    Detectron2. https://github.com/facebookresearch/detectron2

  51. [2021]

    A Study of Multi-Task and Region-Wise Deep Learning for Food Ingredient Recognition.IEEE Transactions on Image Processing30 (2021)

    2021

  52. [2022]

    InAdvances in Neural Information Processing Systems, Vol

    On the Representation Collapse of Sparse Mixture of Experts. InAdvances in Neural Information Processing Systems, Vol. 35. 34600–34613

  53. [2023]

    InProceedings of the IEEE/CVF International Conference on Computer Vision

    Segment Anything. InProceedings of the IEEE/CVF International Conference on Computer Vision. 4015–4026

  54. [2024]

    InProceedings of the European Conference on Computer Vision

    Context-Guided Spatial Feature Reconstruction for Efficient Semantic Segmentation. InProceedings of the European Conference on Computer Vision. Springer Nature Switzerland, 239–255

  55. [2025]

    FoodSAM: Any Food Segmentation.IEEE Transactions on Multimedia27 (2025), 2795–2808

    2025