Pith. sign in

REVIEW 6 major objections 5 minor 35 references

LeMoRe: Learn More Details for Lightweight Semantic Segmentation

T0 review · 6 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper claims LeMoRe reaches 33.5% mIoU on ADE20K with 0.8 GFLOPs and 1.6M parameters, balancing accuracy and compute via explicit and implicit multiview modeling.

desk verdict A plausible architecture, but the headline efficiency trade-off rests on ill-matched baselines and a directly contradicting footnote. read the letter →

arxiv 2505.23093 v1 pith:K3IQ54HI submitted 2025-05-29 cs.CV

Mian Muhammad Naeem Abid , Nancy Mehta , Zongwei Wu , Radu Timofte This is my paper
classification cs.CV
keywords lightweightsemanticsegmentationCartesianencodernestedattentionmultiviewfeaturemodelingefficientvisiontransformerADE20KCityScapesgatedfusion
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 sets out to show that a lightweight semantic segmentation network can match the accuracy of much heavier transformer-based models by replacing a monolithic encoder with a decomposition-modeling-reconstruction scheme. Features are split into explicit Cartesian views, aligned to the height, width, and depth axes, plus implicit views learned through nested attention over query-key pairs. On the ADE20K benchmark the model reports 33.5% mIoU at 0.8 GFLOPs, 1.6M parameters, and 24 ms latency, which the authors claim outperforms comparable lightweight solutions in the efficiency-accuracy trade-off. The same design is shown to transfer to CityScapes, Pascal Context, COCO-Stuff, and object detection.

What carries the argument

The load-bearing mechanism is the Cartesian Encoder: three 1x1 convolutions applied to feature-map permutations along height, width, and depth, yielding transverse, frontal, and lateral views that are summed and refined with depthwise convolutions and channel attention. Complementing it is the Nested Attention module, which forms three query-key-value sets, computes attention maps between every query and every key across the sets, aggregates these into one map, and applies it to all value sets before a feed-forward network. Together they produce the explicit and implicit views that the Gated Fusion Module combines.

What would settle it

Run LeMoRe and, say, LR-ASPP on the same device at 512x512 and measure end-to-end latency and peak memory; if LeMoRe's latency is not substantially lower than the baseline, the efficiency claim is undercut. Also, ablate the nested attention by replacing it with a single global attention of the same parameter budget; if ADE20K mIoU does not drop by roughly 5 points, the claimed mechanism is not the source of the gain.

Watch

Extended reading notes

Core claim

The central claim is that modeling a feature map through multiple complementary views—three predefined Cartesian directions and three learned implicit views—captures local and global context with lower cost than a full transformer encoder. The Cartesian Encoder permutes the feature map along each spatial axis and applies 1x1 convolutions to produce transverse, frontal, and lateral views; the Nested Attention module computes all pairwise query-key attention maps across three sets and aggregates them to weight three value sets. A gated fusion module then recombines the local explicit and global implicit features. The paper reports that each added view contributes a measured mIoU gain (from 27.2% with one view to 33.5% with all three views, channel attention, and nested attention) while GFLOPs stay under 0.81 and parameters at 1.60M on ADE20K. It further reports consistent efficiency advantages on three other segmentation datasets and a detection transfer.

Load-bearing premise

The central trade-off claim assumes that GFLOPs, latency, and parameter counts for all compared baselines are measured under the same protocol, meaning the same input resolution, hardware, and inference framework; in the paper, latency is reported only for LeMoRe and most baseline costs come from their original publications.

Editorial extensions

If this is right

  • If the reported numbers hold, lightweight segmentation can reach roughly 33% mIoU on ADE20K at under 1 GFLOP, a point on the accuracy-efficiency curve that transformer encoders do not currently occupy.
  • The ablation results imply that each additional Cartesian view costs only about 0.02 to 0.04 GFLOPs while adding 0.4 to 1.5 mIoU, making multiview decomposition a cheap source of representational gain.
  • The CityScapes result of 65.0% mIoU at 1.7 GFLOPs suggests the architecture is usable for real-time urban scene parsing, and the ADE20K result supports its use in resource-constrained settings.
  • The detection transfer experiment indicates the learned views generalize as a backbone beyond segmentation, improving mAP over ShuffleNetV2 and MobileNetV3 backbones at similar FLOPs.

Reading between the lines

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

  • The ablation pattern (nested attention alone gives +5.0 mIoU, the Cartesian views add about +1.5) suggests the accuracy gain is driven more by the implicit global-attention term than by the explicit Cartesian decomposition.
  • The same decomposition-modeling-reconstruction pattern could be applied to other dense prediction tasks such as depth estimation, where explicit directional structure may also help.
  • A direct same-hardware latency comparison against LR-ASPP and U-MixFormer would settle whether the 24 ms figure reflects the architecture or favorable implementation; the paper does not provide re-measured latencies for baselines.
  • Because the nested attention computes all cross-products of three query-key sets, its cost grows quadratically with the number of implicit views; the choice of three is a design assumption that may not be optimal for other resolutions.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

6 major / 5 minor

Summary. The paper introduces LeMoRe, a lightweight semantic segmentation architecture that combines explicit Cartesian projections (three 1x1 convolutions applied to spatially permuted feature maps) with implicit views learned through a nested attention mechanism, followed by a gated fusion decoder. The authors report 33.5 mIoU at 0.8 GFLOPs, 1.6M parameters, and 24 ms latency on the ADE20K validation set, together with results on Cityscapes, PASCAL Context, COCO-Stuff, and an object-detection experiment. The central claim is that LeMoRe achieves a superior accuracy/efficiency trade-off compared with lightweight transformer- and CNN-based baselines.

Significance. If the reported trade-off can be confirmed under a common measurement protocol, LeMoRe is a useful contribution to lightweight semantic segmentation: the architecture is simple, transparent, and considerably cheaper than transformer-based lightweight baselines, and the manuscript includes code, per-module ablations, and standard datasets/tooling that aid reproducibility. However, the quantitative basis of the central efficiency claim is currently weakened by inconsistent input-resolution reporting across tables, missing latency/hardware details for most baselines, and several unproven architectural assertions. The paper is therefore a promising candidate for publication after a revision that makes the efficiency comparisons like-for-like.

major comments (6)
  1. [Section 4.1, Table 1] The GFLOPs reporting protocol is internally inconsistent. The table footnote states that GFLOPs are reported for 512x512 input, but the LeMoRe* row, marked as trained with 448x448 input, reports 0.6 GFLOPs; this is exactly 0.8*(448/512)^2, so the number is evidently computed at 448 resolution and would become 0.8 GFLOPs if recomputed at 512, making it identical to the LeMoRe row. Please recompute all FLOPs at one common resolution and clarify what distinguishes LeMoRe* from LeMoRe in cost terms.
  2. [Tables 3-5, Section 4.2] The efficiency numbers in Tables 3-5 mix input resolutions without disclosure. The supplementary states that Cityscapes training uses a 1024x512 crop, so LeMoRe's 1.7 GFLOPs is about twice the 512x512 ADE20K figure; the PSPNet baseline of 423.4 GFLOPs is not plausible at 1024x512 (its ADE20K value at 512x512 is 52.2, implying a roughly 2048x1024 input). The claimed '99.60% reduction' in Section 4.2 is therefore not a like-for-like comparison. Specify the input resolution and source of every FLOPs number in Tables 3-5, or re-measure all baselines under one protocol, and state the hardware and measurement conditions for any latency values.
  3. [Section 4.1, last paragraph] The comparison to LR-ASPP is numerically inconsistent. Against LR-ASPP (MobileNetV3-Large) in Table 1, LeMoRe reduces GFLOPs by 60% (1 - 0.8/2.0) and parameters by 50%, not by 38.5%; the 38.5% figure matches a comparison with MobileNetV3-Large-reduce (1 - 0.8/1.3), for which the mIoU difference is +1.2, not +1.0. Please recompute all relative percentages and explicitly identify the baseline used for each claim.
  4. [Section 3.1, Eq. (1)] The claims that the Cartesian views form an 'orthogonal basis' and 'maximize joint entropy' are unsupported. A spatial permutation followed by a 1x1 convolution does not by itself enforce orthogonality among the resulting views, and no entropy or mutual-information computation is provided. Since these properties are invoked as the motivation for the decomposition, please supply a derivation or remove the claims.
  5. [Section 3.2, Eqs. (2)-(4)] The Nested Attention mechanism is not fully specified. Eq. (3) aggregates pairwise products Qi⊗Kj for i,j=1,2,3 without softmax, scaling, or any normalization, so the resulting matrix is not a valid attention weighting; Eq. (4) then uses this matrix to weight value vectors. Please define the exact normalization and report the computational complexity of the nine pairwise interactions, since the efficiency argument for Nested Attention depends on this.
  6. [Tables 1-5] No repeated runs or error bars are reported. Several central comparisons are small (33.5 vs 33.1 for LR-ASPP; 32.2 vs 32.0 for R-ASPP; ablation increments of 0.4-0.5 mIoU), so run-to-run variation could change the conclusions. Please report mean±std over at least three seeds for the main models and ablations, or provide an equivalent significance assessment.
minor comments (5)
  1. [Abstract and Section 4.1] There are minor language errors: 'comprises of' should be 'comprises' and 'In nutshell' should be 'In a nutshell'.
  2. [Eq. (5)] The symbol ¯⊗ is used in the Gated Fusion formula but is never defined; please define all operators in the equation.
  3. [Section 4.3, Table 6] The object-detection experiment lacks a dataset, input resolution, training schedule, and an explanation of how LeMoRe is adapted to RetinaNet; please add full experimental details or remove this section.
  4. [Introduction, reference [1]] Reference [1] is a human-factors study about automated driving, which appears unrelated to the claim being cited; please replace it with an appropriate computer-vision reference.
  5. [Figure 1(b) and Table 1] The latency values and their units (ms) are given in the table and plot, but no hardware or software stack is specified; please state the measurement conditions for all latency numbers.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: LeMoRe's reported mIoU and efficiency figures are empirical measurements, not quantities derived from their own inputs.

full rationale

The paper's central claim—33.5% mIoU at 0.8 GFLOPs, 1.6M parameters, and 24 ms on ADE20K—is an experimental measurement reported after training. The equations in Section 3 (Eqs. 1–5) define the architecture's forward computation (Cartesian Encoder projections, Nested Attention aggregation, Gated Fusion), but they do not derive or predict any of the reported mIoU numbers, so nothing in the claimed derivation chain reduces to its inputs by construction. No parameter is fitted to a subset of the evaluation data and then renamed as a prediction; the ablations in Table 2 vary architectural components and report empirically measured outcomes, which is standard experimental analysis rather than circular reasoning. The reference list contains no self-citations by the authors (Abid, Mehta, Wu, Timofte), so no self-citation chain is load-bearing, and no uniqueness theorem is invoked anywhere in the text. The strongest concern raised by the reader—that baseline efficiency numbers in Tables 1, 3, 4, and 5 may mix input resolutions and sources (e.g., the LeMoRe* row trained at 448x448 while GFLOPs are said to be for 512x512, and Cityscapes figures plausibly at 1024x512)—is a legitimate question about the comparability and correctness of the efficiency comparison, but it is not a circularity issue: the paper's own numbers are not defined in terms of the baseline numbers, and the baselines are not inputs to any derivation. Under the hard rules, inconsistent measurement protocols constitute a correctness risk, not circular reasoning, so the honest finding is no significant circularity.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The central claim does not rest on fitted mathematical constants. The design choices that matter are the number of Cartesian views, the number of Nested Attention query-key-value sets, and the FFN expansion factor; these are hand-selected and only partially ablated. No new physical entities are introduced: the explicit and implicit views are labels for standard tensor operations, and they have no falsifiable handle outside the final architecture.

free parameters (3)
  • Number of Cartesian views = 3 (transverse, frontal, lateral)
    Chosen by hand; the ablation adds views one at a time, but no search over other counts is reported.
  • Number of Nested Attention QKV sets = 3
    Set in Eq. 2 without ablation; the final architecture depends on this choice.
  • Feed-forward network expansion factor = 2
    The supplementary material states it is used to minimize complexity; no ablation is shown.
assumptions (3)
  • domain assumption Permuting a feature map along height, width, and depth and applying separate 1x1 convolutions yields independent, orthogonal components.
    Section 3.1 asserts orthogonality and 'maximized joint entropy' without proof; the branches share no weights, but their outputs are not mathematically guaranteed to be orthogonal.
  • domain assumption The cascaded depthwise, dilated depthwise, and channel-attention refinement adds only the overhead reported in the parameter and GFLOPs tables.
    The GFLOPs and parameter counts assume these blocks are as cheap as stated; the paper gives no per-component cost breakdown.
  • domain assumption Three query-key-value sets with all-pairs interactions are sufficient to capture the global context needed for segmentation.
    Equation 3 sums nine Q-K interactions, but no ablation varies the number of sets and no theory supports the sufficiency of this pattern.

how reviews work

0 comments
Cite this review

Pith. "Pith review of LeMoRe: Learn More Details for Lightweight Semantic Segmentation." pith.science (2026). https://pith.science/paper/K3IQ54HI

@misc{pith2026250523093,
  author       = {Pith},
  title        = {Pith review of: LeMoRe: Learn More Details for Lightweight Semantic Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/K3IQ54HI}},
  note         = {Machine review of arXiv:2505.23093}
}
read the original abstract

Lightweight semantic segmentation is essential for many downstream vision tasks. Unfortunately, existing methods often struggle to balance efficiency and performance due to the complexity of feature modeling. Many of these existing approaches are constrained by rigid architectures and implicit representation learning, often characterized by parameter-heavy designs and a reliance on computationally intensive Vision Transformer-based frameworks. In this work, we introduce an efficient paradigm by synergizing explicit and implicit modeling to balance computational efficiency with representational fidelity. Our method combines well-defined Cartesian directions with explicitly modeled views and implicitly inferred intermediate representations, efficiently capturing global dependencies through a nested attention mechanism. Extensive experiments on challenging datasets, including ADE20K, CityScapes, Pascal Context, and COCO-Stuff, demonstrate that LeMoRe strikes an effective balance between performance and efficiency.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

35 extracted references · 27 canonical work pages

  1. [1]

    A systematic review and meta-analysis of takeover performance during con- ditionally automated driving,

    B.W. Weaver and P.R. DeLucia, “A systematic review and meta-analysis of takeover performance during con- ditionally automated driving,”Human factors, vol. 64, no. 7, pp. 1227–1260, 2022

    work page 2022

  2. [2]

    Unetformer: A unet-like trans- former for efficient semantic segmentation of remote sensing urban scene imagery,

    L. Wang, R. Li, C. Zhang, S. Fang, C. Duan, X. Meng, and P.M. Atkinson, “Unetformer: A unet-like trans- former for efficient semantic segmentation of remote sensing urban scene imagery,”ISPRS P&RS, vol. 190, pp. 196–214, 2022

    work page 2022

  3. [3]

    U-mixformer: Unet- like transformer with mix-attention for efficient seman- tic segmentation,

    S.K. Yeom and J. von Klitzing, “U-mixformer: Unet- like transformer with mix-attention for efficient seman- tic segmentation,”arXiv preprint arXiv:2312.06272, 2023

    arXiv 2023

  4. [4]

    Scene parsing through ade20k dataset,

    B. Zhou, H. Zhao, X. Puig, S. Fidler, A. Barriuso, and A. Torralba, “Scene parsing through ade20k dataset,” in CVPR, 2017

    2017

  5. [5]

    The cityscapes dataset for semantic urban scene understanding,

    M. Cordts, M. Omran, S. Ramos, T. Rehfeld, M. En- zweiler, R. Benenson, U. Franke, S. Roth, and B. Schiele, “The cityscapes dataset for semantic urban scene understanding,” inCVPR, 2016

    work page 2016

  6. [6]

    The role of con- text for object detection and semantic segmentation in the wild,

    R. Mottaghi, X. Chen, X. Liu, N.G. Cho, S.W. Lee, S. Fidler, R. Urtasun, and A. Yuille, “The role of con- text for object detection and semantic segmentation in the wild,” inCVPR, 2014

    work page 2014

  7. [7]

    Coco-stuff: Thing and stuff classes in context,

    H. Caesar, J. Uijlings, and V . Ferrari, “Coco-stuff: Thing and stuff classes in context,” inCVPR, 2018

    work page 2018

  8. [8]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inCVPR, 2016

    2016

Show all 35 references
  1. [9]

    Mobilenets: Efficient convolutional neural networks for mobile vision applications,

    A.G. Howard, M. Zhu, B. Chen, D. Kalenichenko, W. Wang, T. Weyand, M. Andreetto, and H. Adam, “Mobilenets: Efficient convolutional neural networks for mobile vision applications,”arXiv preprint arXiv:1704.04861, 2017

    2017 arXiv

  2. [10]

    Mobilenetv2: Inverted residuals and linear bottlenecks,

    M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L.C. Chen, “Mobilenetv2: Inverted residuals and linear bottlenecks,” inCVPR, 2018

    2018

  3. [11]

    Shufflenet v2: Practical guidelines for efficient cnn architecture de- sign,

    N. Ma, X. Zhang, H.T. Zheng, and J. Sun, “Shufflenet v2: Practical guidelines for efficient cnn architecture de- sign,” inECCV, 2018

    2018

  4. [12]

    Ghostnet: More features from cheap operations,

    K. Han, Y . Wang, Q. Tian, J. Guo, C. Xu, and C. Xu, “Ghostnet: More features from cheap operations,” in CVPR, 2020

    2020

  5. [13]

    An image is worth 16x16 words: Transformers for image recognition at scale,

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weis- senborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Min- derer, G. Heigold, S. Gelly, et al., “An image is worth 16x16 words: Transformers for image recognition at scale,”arXiv preprint arXiv:2010.11929, 2020

    2010 arXiv

  6. [14]

    Training data-efficient image transformers & distillation through attention,

    H. Touvron, M. Cord, M. Douze, F. Massa, A. Sablay- rolles, and H. J ´egou, “Training data-efficient image transformers & distillation through attention,” inICML, 2021

    2021

  7. [15]

    Levit: a vision transformer in convnet’s clothing for faster inference,

    B. Graham, A. El-Nouby, H. Touvron, P. Stock, A. Joulin, H. J ´egou, and M. Douze, “Levit: a vision transformer in convnet’s clothing for faster inference,” inICCV, 2021

    2021

  8. [16]

    Efficientformer: Vi- sion transformers at mobilenet speed,

    Y . Li, G. Yuan, Y . Wen, J. Hu, G. Evangelidis, S. Tulyakov, Y . Wang, and J. Ren, “Efficientformer: Vi- sion transformers at mobilenet speed,”Adv. Neural Inf. Process. Syst., vol. 35, pp. 12934–12949, 2022

    2022

  9. [17]

    Mobilevit: light- weight, general-purpose, and mobile-friendly vision transformer,

    S. Mehta and M. Rastegari, “Mobilevit: light- weight, general-purpose, and mobile-friendly vision transformer,”arXiv preprint arXiv:2110.02178, 2021

    2021 arXiv

  10. [18]

    Panoptic feature pyramid networks,

    A. Kirillov, R. Girshick, K. He, and P. Doll´ar, “Panoptic feature pyramid networks,” inCVPR, 2019

    2019

  11. [19]

    Pyramid scene parsing network,

    H. Zhao, J. Shi, X. Qi, X. Wang, and J. Jia, “Pyramid scene parsing network,” inCVPR, 2017

    2017

  12. [20]

    Fully convolu- tional networks for semantic segmentation,

    J. Long, E. Shelhamer, and T. Darrell, “Fully convolu- tional networks for semantic segmentation,” inCVPR, 2015

    2015

  13. [21]

    Convmlp: Hierarchical convolutional mlps for vision,

    J. Li, A. Hassani, S. Walton, and H. Shi, “Convmlp: Hierarchical convolutional mlps for vision,” inCVPR, 2023

    2023

  14. [22]

    Encoder-decoder with atrous separable con- volution for semantic image segmentation,

    L.C. Chen, Y . Zhu, G. Papandreou, F. Schroff, and H. Adam, “Encoder-decoder with atrous separable con- volution for semantic image segmentation,” inECCV, 2018

    2018

  15. [23]

    Efficientnet: Rethinking model scal- ing for convolutional neural networks,

    M. Tan and Q. Le, “Efficientnet: Rethinking model scal- ing for convolutional neural networks,” inICML, 2019

    2019

  16. [24]

    Pem: Prototype- based efficient maskformer for image segmentation,

    N. Cavagnero, G. Rosi, C. Cuttano, F. Pistilli, M. Ci- ccone, G. Averta, and F. Cermelli, “Pem: Prototype- based efficient maskformer for image segmentation,” in CVPR, 2024

    2024

  17. [25]

    Object-contextual representations for semantic segmentation,

    Y . Yuan, X. Chen, and J. Wang, “Object-contextual representations for semantic segmentation,” inECCV, 2020

    2020

  18. [26]

    Segformer: Simple and efficient design for semantic segmentation with transformers,

    E. Xie, W. Wang, Z. Yu, A. Anandkumar, J.M. Alvarez, and P. Luo, “Segformer: Simple and efficient design for semantic segmentation with transformers,”Adv. Neural Inf. Process. Syst., vol. 34, pp. 12077–12090, 2021

    2021

  19. [27]

    Feed- former: Revisiting transformer decoder for efficient se- mantic segmentation,

    J.h. Shim, H. Yu, K. Kong, and S.J. Kang, “Feed- former: Revisiting transformer decoder for efficient se- mantic segmentation,” inAAAI, 2023

    2023

  20. [28]

    Segnext: Rethinking convolutional attention design for semantic segmentation,

    M.H. Guo, C.Z. Lu, Q. Hou, Z. Liu, M.M. Cheng, and S.M. Hu, “Segnext: Rethinking convolutional attention design for semantic segmentation,”Adv. Neural Inf. Pro- cess. Syst., vol. 35, pp. 1140–1156, 2022

    2022

  21. [29]

    Hr-nas: Searching efficient high-resolution neural architectures with lightweight transformers,

    M. Ding, X. Lian, L. Yang, P. Wang, X. Jin, Z. Lu, and P. Luo, “Hr-nas: Searching efficient high-resolution neural architectures with lightweight transformers,” in CVPR, 2021

    2021

  22. [30]

    Searching for mobilenetv3,

    A. Howard, M. Sandler, G. Chu, L.C. Chen, B. Chen, M. Tan, W. Wang, Y . Zhu, R. Pang, V . Vasudevan, et al., “Searching for mobilenetv3,” inICCV, 2019

    2019

  23. [31]

    Enet: A deep neural network architecture for real-time semantic segmentation,

    A. Paszke, A. Chaurasia, S. Kim, and E. Culurciello, “Enet: A deep neural network architecture for real-time semantic segmentation,”arXiv:1606.02147, 2016

    2016 arXiv

  24. [32]

    Metaseg: Metaformer-based global contexts-aware network for efficient semantic segmentation,

    B. Kang, S. Moon, Y . Cho, H. Yu, and S.J. Kang, “Metaseg: Metaformer-based global contexts-aware network for efficient semantic segmentation,” inWACV, 2024

    2024

  25. [33]

    Topformer: Token pyramid trans- former for mobile semantic segmentation,

    W. Zhang, Z. Huang, G. Luo, T. Chen, X. Wang, W. Liu, G. Yu, and C. Shen, “Topformer: Token pyramid trans- former for mobile semantic segmentation,” inCVPR, 2022

    2022

  26. [34]

    MMSegmentation: Openmm- lab semantic segmentation toolbox and bench- mark,

    M. Contributors, “MMSegmentation: Openmm- lab semantic segmentation toolbox and bench- mark,”https://github.com/open-mmlab/ mmsegmentation, 2020

    2020

  27. [35]

    Imagenet: A large-scale hierarchical im- age database,

    J. Deng, W. Dong, R. Socher, L.J. Li, K. Li, and L. Fei-Fei, “Imagenet: A large-scale hierarchical im- age database,” inCVPR, 2009. — Supplementary Material — ImageNet Pre-training To ensure a fair comparison, we initialize the LeMoRe model with pre-trained parameters from Ima...

    2009

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.