Pith. sign in

REVIEW 4 major objections 5 minor 54 references

SAG-ViT: A Scale-Aware, High-Fidelity Patching Approach with Graph Attention for Vision Transformers

T0 review · 4 major / 5 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 SAG-ViT patches CNN feature maps instead of raw images, routes the patches through a similarity-weighted graph and a GAT, and reports top F1 scores on six benchmarks at lower memory than ViT-L.

desk verdict A plausible mashup of known components whose central empirical claim is undercut by internally inconsistent baselines and an impossible parameter table. read the letter →

arxiv 2411.09420 v3 pith:NBKLQUME submitted 2024-11-14 cs.CV cs.AIcs.LG

Shravan Venkatraman , Jaskaran Singh Walia , Joe Dhanith P R This is my paper · ORCID
classification cs.CVcs.AIcs.LG
keywords visiontransformersgraphattentionnetworksmulti-scalefeaturesimageclassificationfeaturemappatchingEfficientNetV2k-connectivitytokenization
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

Image classifiers based on Vision Transformers normally cut the raw image into patches and let self-attention find the structure. SAG-ViT proposes instead to let a CNN backbone first produce multi-scale feature maps, patch those maps, and then treat the patches as nodes of a k-connectivity graph with similarity-weighted edges that a Graph Attention Network refines before a Transformer encoder reads the result. The paper reports F1 scores of 0.9574 on CIFAR-10, 0.9958 on GTSRB, 0.9861 on NCT-CRC-HE-100K, 0.9772 on PlantVillage, 0.9549 on NWPU-RESISC45, and 0.9615 on the Underwater Trash Dataset, and argues that this graph-attended feature-map patching captures multi-scale semantics more cheaply than raw-image patching. If the claim holds, a comparatively light CNN-plus-graph front end can give a Transformer the local structure it otherwise pays much more compute to learn, at lower memory cost than a large ViT.

What carries the argument

The load-bearing mechanism is the patching-plus-graph front end. Feature maps $F$ from EfficientNetV2 are divided into non-overlapping $4\times4$ patches, each vectorized to a node; edges are drawn between nodes in a $k$-nearest spatial neighborhood and weighted by $\exp(-\|x_u - x_v\|^2 / \sigma^2)$, making the adjacency sparse. A multi-head Graph Attention Network recomputes attention coefficients over these edges and concatenates head outputs to refine node embeddings, which are then summed with positional encodings and passed through a standard Transformer encoder; a global mean-pooled token feeds an MLP classifier. The $k$-connectivity keeps the graph sparse, and patching at the feature-map level means fewer, semantically richer tokens than patching the input image at the same window size.

What would settle it

Re-run the paper's CIFAR-10 experiment with the same EfficientNetV2-to-Transformer stack in the 'Backbone + Transformer (No GAT)' configuration under the stated training protocol (128 epochs, Adam, cosine schedule, batch size 128) and compare its F1 to the reported 0.1428; if the no-GAT baseline instead lands near 0.759, as the paper's own ablation Table 9 reports for the same configuration, then the +570% improvement and the attribution of the gain to the GAT are not supported by the evidence as presented.

Watch

Extended reading notes

Core claim

On its own terms, the discovery is that the right place to spend modeling effort in a ViT is the tokenization step. Instead of linearizing raw pixels, SAG-ViT takes strided feature maps from EfficientNetV2, divides them into non-overlapping patches of size 4-by-4, connects the vectorized patches through a k-connectivity graph whose edges carry Gaussian similarity weights, and lets a multi-head GAT refine the node embeddings before the Transformer encoder performs global self-attention. The paper reports that this pipeline consistently outperforms its own CNN-to-ViT baselines across all six datasets, with the largest margins exactly where the plain baselines score near random chance, and that the full model uses less RAM and GPU memory than ViT-L. The authors read the result as evidence that feature-map patching preserves semantic context that raw-image patching loses, and that the GAT provides the local dependency modeling the Transformer alone lacks.

Load-bearing premise

The load-bearing premise is that the no-GAT CNN-to-ViT baselines in Tables 1-6 are faithful, correctly trained controls; the paper's own tables strain this premise by reporting F1 = 0.759 for 'Backbone + Transformer (No GAT)' in the ablation and F1 = 0.1428 for the same configuration in the main CIFAR-10 comparison.

Editorial extensions

If this is right

  • If the reported numbers hold, tokenizing CNN feature maps rather than raw pixels becomes a practical recipe for cutting a ViT's memory footprint: on CIFAR-10 SAG-ViT is reported at 33.12% GPU VRAM versus 81.87% for ViT-L, with a higher F1.
  • The GAT stage is the component credited with local context: the ablation shows the Backbone + Transformer (No GAT) configuration falling to near-random F1 on GTSRB, PlantVillage, and NWPU-RESISC45, while Backbone + GAT keeps strong scores.
  • The same fixed recipe, 224-by-224 input, 4-by-4 patch, and k-connectivity graph, is claimed to reach high F1 on six different domains, so the pipeline would need no per-dataset re-engineering to transfer.
  • Throughput stays around 370 images/s in the reported comparisons, which suggests the added graph-attention stage does not exact an obvious throughput penalty relative to the plain backbone-plus-transformer baseline.

Reading between the lines

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

  • Editorial inference: because the k-connectivity graph is a fixed sparse mask, a direct test the paper does not run is replacing the GAT with masked self-attention over the same graph; that comparison would isolate what the GAT's attention coefficients add beyond the graph connectivity itself.
  • Editorial inference: since the backbone stride shrinks the feature map before patching, the token count grows far more slowly with input resolution than a raw-image ViT's token count, so the efficiency advantage should widen at higher resolutions than the 224-by-224 used here.
  • Editorial inference: the Euclidean similarity weight in the adjacency uses raw CNN feature vectors; a learned projection of patch features before edge weighting could make the graph more class-discriminative, but the paper does not explore that variant.
  • Editorial inference: if the method transfers as reported, the feature-map patcher could be grafted onto other ViT-based tasks like detection or segmentation that already compute multi-scale feature maps, but the paper only demonstrates classification.
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 / 5 minor

Summary. The paper proposes SAG-ViT, a hybrid image classification architecture that passes images through an EfficientNetV2 backbone, patches the resulting feature maps, constructs k-connectivity graphs with similarity-weighted edges, refines node embeddings with a Graph Attention Network (GAT), and finally processes the embeddings with a Transformer encoder. The authors report high F1 scores on six datasets (CIFAR-10, GTSRB, NCT-CRC-HE-100K, PlantVillage, NWPU-RESISC45, UTD) and claim that SAG-ViT consistently outperforms existing backbone-assisted transformer-based approaches while using less memory than ViT-L. The experimental evidence consists mainly of comparisons to the authors' own [CNN->ViT] stacks with and without a GAT, together with hardware utilization tables, loss landscape visualizations, UMAP plots, and an ablation study.

Significance. If the architecture and the reported comparisons were reliable, the combination of CNN feature-map patching with graph attention before a Transformer encoder could be a useful efficiency-preserving tokenization strategy. The paper covers six diverse datasets, reports throughput and memory usage, and makes code and weights publicly available, which are positive features. However, the validity of the central claims rests on the internal controls, and those controls are inconsistent in ways that undermine the headline F1 gains and the efficiency claims. Without corrected experiments and comparisons to published methods, the paper's main conclusions are not supported.

major comments (4)
  1. [Section 4.1, Tables 1 and 9] The no-GAT control is internally inconsistent. Table 1 reports EfficientNetV2t (EfficientNetV2->ViT without GAT) as F1=0.1428 on CIFAR-10, while Table 9's 'Backbone + Transformer (No GAT)', described in Section 4.4 as the same EfficientNetV2 backbone without GAT, reports F1=0.759 on the same dataset under the same training settings. The two rows also disagree on PlantVillage (0.2876 vs 0.033), NWPU-RESISC45 (0.0228 vs 0.007), GTSRB (0.1246 vs 0.009), NCT-CRC-HE-100K (0.4352 vs 0.169), and UTD (0.4938 vs 0.390). At least one of these two sets of numbers cannot be correct, so the headline +570% improvement on CIFAR-10 and the attribution of the gain to the GAT are unsupported until the discrepancy is resolved.
  2. [Table 8] The reported parameter counts decrease when a GAT is added, which is not possible for the described architecture. For example, EfficientNetV2t goes from 9.76M to 6.39M parameters, ShuffleNetV2t from 3.73M to 1.26M, and SqueezeNet from 2.99M to 0.75M. Since the GAT introduces a learned transformation matrix W and attention vector a (Eq. 3), adding it cannot reduce the total parameter count. This indicates a counting or configuration error and invalidates the parameter-reduction efficiency claim.
  3. [Section 4.1 and Introduction] The central claim that SAG-ViT 'consistently outperforms existing backbone-assisted transformer-based approaches' is not supported by the experiments. Tables 1-6 compare only against the authors' own [CNN->ViT] stacks with and without GAT; there is no comparison with any published backbone-assisted transformer method such as CrossViT, CvT, MobileViT, LeViT, or BoTNet. Without such comparisons, the claim of superiority over existing approaches is untested.
  4. [Section 4.3, Table 7] The memory-efficiency claim is inconsistent with the data on several datasets. SAG-ViT uses more GPU VRAM than ViT-L on GTSRB (36.38% vs 33.72%), NCT-CRC-HE-100K (37.32% vs 33.21%), and PlantVillage (39.32% vs 35.98%), although the introduction and abstract state that the method achieves significantly lower GPU VRAM usage than ViT-L. The claim needs qualification or the measurements need to be rechecked.
minor comments (5)
  1. [Section 3.2, Algorithm 2] Algorithm 2 calls 'PatchToGraph(Pn, ...)' but Algorithm 1 defines the function 'GENERATE_GRAPH'; the pseudocode should use consistent names. The variable F is also used both for the feature map and as a dimension in Algorithm 2, which makes the notation ambiguous.
  2. [Section 4.4, Table 9] The ablation table reports RAM and GPU VRAM in GB, while the main efficiency table (Table 7) reports percentages. Please state the reference totals so the two tables can be compared directly.
  3. [Figure 4 and Figure 5] The UMAP and token-correlation visualizations are qualitative; adding quantitative metrics, such as silhouette scores or average pairwise correlation, would make the claimed representational advantages testable.
  4. [Tables 1-6] No standard deviations or repeated-run statistics are reported. Several performance differences are small (e.g., 1.49% on GTSRB and 4.04% on NCT-CRC-HE-100K), so single-run F1 values are insufficient to establish the ordering of methods.
  5. [Section 4, Training Settings] The paper says it 'primarily adopts' the training settings of [41] but does not specify which settings are modified; please list the optimizer, learning-rate schedule, augmentations, and backbone initialization explicitly.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation is present: SAG-ViT is an empirical architecture paper whose equations and evaluations do not reduce to their own inputs by construction.

full rationale

This paper contains no formal derivation chain that could collapse into its inputs. The method is defined by standard components: CNN feature-map patching, k-connectivity graph construction with similarity-weighted edges, a GAT, and a transformer encoder (Eqs. 1-7). None of these equations re-introduces a target result as a premise, and the reported F1 scores are measured on test splits rather than quantities reconstructed from fitted parameters, so the 'fitted input called prediction' pattern does not apply. The only self-citations are references [45] and [46] by co-author Walia, which are used purely as the source of the UTD dataset; they are external data references and are not load-bearing for any architectural claim. The paper's central comparison is weakened by internal inconsistencies: the same 'Backbone + Transformer (No GAT)' configuration reports F1 = 0.1428 in Table 1 but F1 = 0.759 in Table 9 on CIFAR-10, and Table 8 reports parameter-count reductions after adding GAT, which is implausible since a GAT introduces learnable weights. These are serious correctness and reproducibility concerns, but they are not circular reasoning: no numerical result is shown to equal another by definition or by the paper's own equations. Accordingly, no circular step is flagged.

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

The method rests on several unproved modeling choices: that CNN feature maps are better than raw images for patching, that a fixed spatial-neighbor graph with Gaussian similarity edges is the right relational inductive bias, and that hand-picked hyperparameters (patch size, sigma, k) are sufficient. No new physical or mathematical entities are introduced.

free parameters (4)
  • sigma (similarity decay)
    Appears in Eq. (1) as the denominator of exp(-||xu-xv||^2/sigma^2); no value or selection procedure is reported, yet it controls the adjacency weights.
  • patch size = (4,4)
    Chosen after trying (2,2) and (6,6); the choice is load-bearing for graph density, token count, and the reported accuracy-efficiency trade-off.
  • k (neighborhood size) = 8 (3x3 window)
    Algorithm 1 connects the 8 spatial neighbors, while Eq. (2) phrases k as k-nearest spatial neighbors; the exact k used in experiments is not reported.
  • GAT and transformer dimensions
    Layer counts, hidden widths, number of heads, and d_model are not stated, yet they determine the reported F1 and efficiency numbers.
assumptions (5)
  • domain assumption Pretrained CNN feature maps retain richer semantic and multi-scale information than raw-image patching.
    Section 3.1 motivates patching feature maps from EfficientNetV2; the entire architecture depends on this premise, which is not independently established here.
  • domain assumption Grid-based spatial adjacency and Gaussian feature similarity in Eq. (1) define a graph that captures relationships needed for classification.
    Section 3.2 constructs edges from spatial neighbors and exp(-||xu-xv||^2/sigma^2); no analysis or ablation shows this weighting is superior to other edge definitions.
  • domain assumption Appending GAT before the transformer improves local modeling without harming global dependencies.
    Section 3.3 claims GAT refines node embeddings; the only support is the near-random baseline comparison in Tables 1-6.
  • domain assumption Flat loss landscapes imply good generalization.
    Section 4.2 interprets loss landscape shapes as evidence of generalization, an assumption not proven by the visualizations.
  • ad hoc to paper The chosen patch size (4,4) balances efficiency and fidelity.
    Section 1 reports that (2,2) was too costly and (6,6) lost detail, but no ablation table is given for this decision.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SAG-ViT: A Scale-Aware, High-Fidelity Patching Approach with Graph Attention for Vision Transformers." pith.science (2026). https://pith.science/paper/NBKLQUME

@misc{pith2026241109420,
  author       = {Pith},
  title        = {Pith review of: SAG-ViT: A Scale-Aware, High-Fidelity Patching Approach with Graph Attention for Vision Transformers},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NBKLQUME}},
  note         = {Machine review of arXiv:2411.09420}
}
read the original abstract

Vision Transformers (ViTs) have redefined image classification by leveraging self-attention to capture complex patterns and long-range dependencies between image patches. However, a key challenge for ViTs is efficiently incorporating multi-scale feature representations, which is inherent in convolutional neural networks (CNNs) through their hierarchical structure. Graph transformers have made strides in addressing this by leveraging graph-based modeling, but they often lose or insufficiently represent spatial hierarchies, especially since redundant or less relevant areas dilute the image's contextual representation. To bridge this gap, we propose SAG-ViT, a Scale-Aware Graph Attention ViT that integrates multi-scale feature capabilities of CNNs, representational power of ViTs, graph-attended patching to enable richer contextual representation. Using EfficientNetV2 as a backbone, the model extracts multi-scale feature maps, dividing them into patches to preserve richer semantic information compared to directly patching the input images. The patches are structured into a graph using spatial and feature similarities, where a Graph Attention Network (GAT) refines the node embeddings. This refined graph representation is then processed by a Transformer encoder, capturing long-range dependencies and complex interactions. We evaluate SAG-ViT on benchmark datasets across various domains, validating its effectiveness in advancing image classification tasks. Our code and weights are available at https://github.com/shravan-18/SAG-ViT.

Figures

Figures reproduced from arXiv: 2411.09420 by the authors.

Figure 1
Figure 1. Resource usage and computational complexity com [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Visualization of the patch generation and graph construction pipeline in SAG-ViT. [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. An illustration of our proposed SAG-ViT architecture for learning scale-aware, high-fidelity features with graph attention. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Feature embeddings projected into a 2D space using UMAP for DeiT, Vanilla ViT, and SAG-ViT (Proposed). [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Token-token correlation matrix comparison of SAG-ViT with DeiT and vanilla ViT. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: F1 Score and Throughput comparison of SAG-ViT and [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Comparison of loss landscapes for the final convolutional, graph attention, and Transformer layers, showcasing parameter [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Ablation study results for CIFAR-10, GTSRB, NCT-CRC-HE-100K, NWPU-RESISC45, PlantVillage, and UTD datasets. [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

54 extracted references · 49 canonical work pages

  1. [1]

    Uni4eye++: A general masked image modeling multi-modal pre-training framework for ophthalmic image classification and segmentation

    Zhiyuan Cai, Li Lin, Huaqing He, Pujin Cheng, and Xiaoy- ing Tang. Uni4eye++: A general masked image modeling multi-modal pre-training framework for ophthalmic image classification and segmentation. IEEE Transactions on Med- ical Imaging, 43(12):4419–4429, 2024. 13

    work page 2024

  2. [2]

    End-to- end object detection with transformers, 2020

    Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End-to- end object detection with transformers, 2020. 13

    work page 2020

  3. [3]

    Crossvit: Cross-attention multi-scale vision transformer for image classification

    Chun-Fu Chen, Quan Fan, and Rajeev Panda. Crossvit: Cross-attention multi-scale vision transformer for image classification. 2021. 2, 3

    work page 2021

  4. [4]

    Chen et al

    L. Chen et al. Hybrid-structure-oriented transformer for arm musculoskeletal ultrasound segmentation. In Medical Image Computing and Computer Assisted Intervention – MICCAI 2024, pages 621—-631. Springer, Cham, 2024. 13

    work page 2024

  5. [5]

    Ef- ficient decoder-free object detection with transformers, 2022

    Peixian Chen, Mengdan Zhang, Yunhang Shen, Kekai Sheng, Yuting Gao, Xing Sun, Ke Li, and Chunhua Shen. Ef- ficient decoder-free object detection with transformers, 2022. 13

    work page 2022

  6. [6]

    Yinpeng et al. Chen. Drop an octave: Reducing spatial re- dundancy in convolutional neural networks with octave con- volution. In 2019 IEEE/CVF International Conference on Computer Vision (ICCV), pages 3434–3443. IEEE, 2019. 3

    work page 2019

  7. [7]

    Yinpeng et al. Chen. Mobile-former: Bridging mobilenet and transformer, 2021. 3

    work page 2021

  8. [8]

    Remote sens- ing image scene classification: Benchmark and state of the art

    Gong Cheng, Junwei Han, and Xiaoqiang Lu. Remote sens- ing image scene classification: Benchmark and state of the art. Proceedings of the IEEE, 105(10):1865–1883, 2017. 2, 7

    work page 2017

Show all 54 references
  1. [9]

    Dosovitskiy

    Alexey et al. Dosovitskiy. An image is worth 16x16 words: Transformers for image recognition at scale. 2020. 1

    2020

  2. [10]

    Large- scale learnable graph convolutional networks

    Hongyang Gao, Zhengyang Wang, and Shuiwang Ji. Large- scale learnable graph convolutional networks. 2018. 3

    2018

  3. [11]

    Benjamin et al. Graham. Levit: a vision transformer in con- vnet’s clothing for faster inference, 2021. 3

    2021

  4. [12]

    Jiang et al. Guo. Cmt: Convolutional neural networks meet vision transformers. 2021. 3

    2021

  5. [13]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. pages 770– 778, 2016. 7, 8, 9

    2016

  6. [14]

    Densely connected convolutional net- works

    Gao Huang, Zhuang Liu, Laurens Van Der Maaten, and Kil- ian Q Weinberger. Densely connected convolutional net- works. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 4700–4708, 2017. 7, 8, 9

    2017

  7. [15]

    Polarformer: A transformer-based method for multi-lesion segmentation in intravascular oct

    Zhili Huang, Jingyi Sun, Yifan Shao, Zixuan Wang, Su Wang, Qiyong Li, Jinsong Li, and Qian Yu. Polarformer: A transformer-based method for multi-lesion segmentation in intravascular oct. IEEE Transactions on Medical Imaging , 43(12):4190–4199, 2024. 13

    2024

  8. [16]

    David. P. Hughes and Marcel Salathe. An open access repos- itory of images on plant health to enable the development of mobile disease diagnostics, 2016. 2, 7

    2016

  9. [17]

    Iandola, Song Han, Matthew W

    Forrest N. Iandola, Song Han, Matthew W. Moskewicz, Khalid Ashraf, William J. Dally, and Kurt Keutzer. Squeezenet: Alexnet-level accuracy with 50x fewer parame- ters and ¡0.5mb model size, 2016. 7, 8, 9

    2016

  10. [18]

    Nct-crc-he: Not all histopathological datasets are equally useful, 2024

    Andrey Ignatov and Grigory Malivenko. Nct-crc-he: Not all histopathological datasets are equally useful, 2024. 2, 7

    2024

  11. [19]

    Andrew et al. Jaegle. Perceiver: General perception with iterative attention. 2021. 3

    2021

  12. [20]

    Adam: A method for stochastic optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 ,

    arXiv

  13. [21]

    Learning multiple layers of features from tiny images

    Alex Krizhevsky. Learning multiple layers of features from tiny images. Technical report, 2009. 2, 7

    2009

  14. [22]

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton. Imagenet classification with deep convolutional neural net- works. Commun. ACM, 60(6):84–90, 2017. 1, 3

    2017

  15. [23]

    Lee, J.S

    J.B. Lee, J.S. Kim, and H.G. Lee. Covid19 to pneumo- nia: Multi region lung severity classification using cnn trans- former position-aware feature encoding network. In Medi- cal Image Computing and Computer Assisted Intervention – MICCAI 2024, pages 472–481. Springer, Cham, 2024. 13

    2024

  16. [24]

    Feature pyramid networks for object detection, 2016

    Tsung-Yi Lin, Piotr Doll ´ar, Ross Girshick, Kaiming He, Bharath Hariharan, and Serge Belongie. Feature pyramid networks for object detection, 2016. 3

    2016

  17. [25]

    Feature pyramid networks for object detection, 2017

    Tsung-Yi Lin, Piotr Doll ´ar, Ross Girshick, Kaiming He, Bharath Hariharan, and Serge Belongie. Feature pyramid networks for object detection, 2017. 2

    2017

  18. [26]

    Ze et al. Liu. Swin transformer: Hierarchical vision trans- former using shifted windows. 2021. 3 13

    2021

  19. [27]

    Understanding the effective receptive field in deep convolu- tional neural networks

    Wenjie Luo, Yujia Li, Raquel Urtasun, and Richard Zemel. Understanding the effective receptive field in deep convolu- tional neural networks. 2017. 1

    2017

  20. [28]

    Shufflenet v2: Practical guidelines for efficient cnn architec- ture design

    Ningning Ma, Xiangyu Zhang, Hai-Tao Zheng, and Jian Sun. Shufflenet v2: Practical guidelines for efficient cnn architec- ture design. In Proceedings of the European Conference on Computer Vision (ECCV), pages 116–131, 2018. 7, 8, 9

    2018

  21. [29]

    Mobilevit: Light- weight, general-purpose, and mobile-friendly vision trans- former, 2021

    Sachin Mehta and Mohammad Rastegari. Mobilevit: Light- weight, general-purpose, and mobile-friendly vision trans- former, 2021. 3

    2021

  22. [30]

    Mark et al. Sandler. Mobilenetv2: Inverted residuals and linear bottlenecks, 2018. 3, 7, 8, 9

    2018

  23. [31]

    Very deep convo- lutional networks for large-scale image recognition

    Karen Simonyan and Andrew Zisserman. Very deep convo- lutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556, 2014. 7, 8, 9

    2014 arXiv

  24. [32]

    Vidt: An efficient and effective fully transformer-based object detector, 2021

    Hwanjun Song, Deqing Sun, Sanghyuk Chun, Varun Jam- pani, Dongyoon Han, Byeongho Heo, Wonjae Kim, and Ming-Hsuan Yang. Vidt: An efficient and effective fully transformer-based object detector, 2021. 13

    2021

  25. [33]

    Srinivas

    Aravind et al. Srinivas. Bottleneck transformers for visual recognition, 2021. 3

    2021

  26. [34]

    Johannes Stallkamp, Marc Schlipsing, Jan Salmen, and Christian Igel. Man vs. computer: Benchmarking machine learning algorithms for traffic sign recognition. Neural Net- works, 32:323–332, 2012. 2, 7

    2012

  27. [35]

    Going deeper with convolutions

    Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, and Andrew Rabinovich. Going deeper with convolutions. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 1–9, 2015. 7, 8, 9

    2015

  28. [36]

    Rethinking the inception archi- tecture for computer vision

    Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jon Shlens, and Zbigniew Wojna. Rethinking the inception archi- tecture for computer vision. InProceedings of the IEEE Con- ference on Computer Vision and Pattern Recognition, pages 2818–2826, 2016. 7, 8, 9

    2016

  29. [37]

    Mingxing Tan and Quoc V . Le. Efficientnet: Rethinking model scaling for convolutional neural networks, 2020. 7, 8, 9

    2020

  30. [38]

    Mingxing Tan and Quoc V . Le. Efficientnetv2: Smaller mod- els and faster training. 2021. 2

    2021

  31. [39]

    Category feature transformer for semantic segmentation, 2023

    Quan Tang, Chuanjian Liu, Fagui Liu, Yifan Liu, Jun Jiang, Bowen Zhang, Kai Han, and Yunhe Wang. Category feature transformer for semantic segmentation, 2023. 13

    2023

  32. [40]

    X. Tian, N. Anantrasirichai, L. Nicholson, and A. Achim. Tagat: Topology-aware graph attention network for multi- modal retinal image fusion. In Medical Image Comput- ing and Computer Assisted Intervention – MICCAI 2024 . Springer, Cham, 2024. 13

    2024

  33. [41]

    Training data-efficient image transformers & distillation through at- tention

    Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Herv ´e J´egou. Training data-efficient image transformers & distillation through at- tention. arXiv preprint arXiv:2012.12877, 2020. 7

    2012 arXiv

  34. [42]

    Training data-efficient image transformers & distillation through at- tention

    Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Herv ´e J´egou. Training data-efficient image transformers & distillation through at- tention. 2020. 3

    2020

  35. [43]

    Ashish et al. Vaswani. Attention is all you need. 2017. 1

    2017

  36. [44]

    Graph at- tention networks

    Petar Veli ˇckovi´c, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Li `o, and Yoshua Bengio. Graph at- tention networks. 2017. 2, 3

    2017

  37. [45]

    Deep learning inno- vations for underwater waste detection: An in-depth analy- sis, 2024

    Jaskaran Singh Walia and Pavithra L K. Deep learning inno- vations for underwater waste detection: An in-depth analy- sis, 2024. 2, 7

    2024

  38. [46]

    Optimized custom dataset for efficient detection of underwater trash

    Jaskaran Singh Walia and Karthik Seemakurthy. Optimized custom dataset for efficient detection of underwater trash. In Towards Autonomous Robotic Systems , pages 292–303, Cham, 2023. Springer Nature Switzerland. 2, 7

    2023

  39. [47]

    Non-local neural networks

    Xiaolong Wang, Ross Girshick, Abhinav Gupta, and Kaim- ing He. Non-local neural networks. In 2018 IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 7794–7803. IEEE, 2018. 3

    2018

  40. [48]

    Haiping et al. Wu. Cvt: Introducing convolutions to vision transformers. 2021. 3

    2021

  41. [49]

    Zonghan Wu, Shirui Pan, Fengwen Chen, Guodong Long, Chengqi Zhang, and Philip S. Yu. A comprehensive survey on graph neural networks.IEEE Transactions on Neural Net- works and Learning Systems, 2019. 2, 3

    2019

  42. [50]

    Multi-label chest x-ray image classification with single positive labels

    Jiayin Xiao, Si Li, Tongxu Lin, Jian Zhu, Xiaochen Yuan, David Dagan Feng, and Bin Sheng. Multi-label chest x-ray image classification with single positive labels. IEEE Trans- actions on Medical Imaging, 43(12):4404–4418, 2024. 13

    2024

  43. [51]

    Alvarez, and Ping Luo

    Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, and Ping Luo. Segformer: Simple and ef- ficient design for semantic segmentation with transformers,

  44. [52]

    Li et al. Yuan. Tokens-to-token vit: Training vision trans- formers from scratch on imagenet. 2021. 3

    2021

  45. [53]

    Q. Zhou, H. Zou, Z. Wang, H. Jiang, and Y . Wang. Refining intraocular lens power calculation: A multi-modal frame- work using cross-layer attention and effective channel at- tention. In Medical Image Computing and Computer As- sisted Intervention – MICCAI 2024, page 754–763. ...

    2024

  46. [54]

    Yixin et al. Zhu. A survey on graph structure learning: Progress and opportunities. 2021. 3 14

    2021

Pith tools

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