Pith. sign in

REVIEW 3 major objections 8 minor 3 cited by

DCFormer: Efficient 3D Vision-Language Modeling with Decomposed Convolutions

T0 review · 3 major / 8 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read The paper claims that factorizing each 3D convolution into three parallel 1D depthwise convolutions cuts the per-layer cost from $Ck^3$ to $3Ck$ parameters and from $2CHWDk^3$ to $6CHWDk$ FLOPs while matching or beating larger encoders on…

desk verdict The efficiency gain is real and the architecture is clearly described, but the paper never runs the one control experiment that would support its claim that decomposed convolutions preserve 3D spatial information. read the letter →

arxiv 2502.05091 v2 pith:LXS3VP5W submitted 2025-02-07 cs.CV

Gorkem Can Ates , Yu Xin , Kuang Gong , Wei Shao This is my paper · ORCID
classification cs.CV
keywords 3Dvision-languagemodeldecomposedconvolutiondepthwiseseparableCT-RATEzero-shotpathologydetectionCLIPmedicalimageanalysisefficientdeeplearning
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 tries to establish that a single architectural substitution—replacing each 3D convolution with three parallel 1D depthwise convolutions along depth, height, and width—is enough to make a 3D vision-language encoder for medical CT efficient without sacrificing accuracy. The authors build a CLIP-style model whose image encoder is a hierarchical stack of such decomposed-convolution blocks, and train it on 50,188 chest CT volumes paired with radiology reports. In zero-shot and fine-tuned detection of 18 abnormalities, and in image-text retrieval, their smallest encoders match or beat much heavier baselines while using far fewer parameters and FLOPs. If the claim holds, the same encoder design could become a cheap drop-in image tower for volumetric medical vision-language systems.

What carries the argument

The load-bearing mechanism is the Decomposed Convolution, defined in Eq. 7 as $$X' = X + \mathrm{Norm}_h(X \ast_h) + \mathrm{Norm}_w(X \ast_w) + \mathrm{Norm}_d(X \ast_d),$$ where $X \ast_h = \mathrm{DWConv}_{k\times1\times1}(X)$, $X \ast_w = \mathrm{DWConv}_{1\times k\times1}(X)$, and $X \ast_d = \mathrm{DWConv}_{1\times1\times k}(X)$ are depthwise convolutions along height, width, and depth, each followed by its own normalization. The block inherits the MetaFormer layout—normalization, token mixer, channel MLP, residual connection—with the decomposed convolution in place of self-attention. This is the component that turns $Ck^3$ parameters into $3Ck$ and $2CHWDk^3$ FLOPs into $6CHWDk$, which is what makes large kernels (here 7) affordable in deep 3D networks.

What would settle it

Run the same CLIP training with everything identical except that the token mixer is a full 3D depthwise convolution matched in FLOPs or parameters, and compare zero-shot F1 on the 18 CT-RATE abnormalities; a clear win for the full mixer, or a sharp drop when any one of the three axis branches is removed, would show the additive decomposition is losing information that the reported tables cannot resolve.

Watch

Extended reading notes

Core claim

The central claim is that a 3D depthwise convolution can be replaced, as a token mixer, by three parallel 1D depthwise convolutions—one along each spatial axis—and still serve as the spatial backbone of a CLIP-style vision-language model. The replacement changes the per-layer cost from $Ck^3$ parameters and $2CHWDk^3$ FLOPs to $3Ck$ parameters and $6CHWDk$ FLOPs, turning cubic growth in kernel size into linear growth. On CT-RATE, DCFormer variants report the highest F1 scores in both zero-shot and fine-tuned detection of 18 chest abnormalities among the compared encoders, while the naïve variant uses 5.85 million parameters against CT-ViT's 101.1 million. The authors attribute the result to large-kernel decomposed convolutions giving broad spatial context inside a hierarchical, multi-scale encoder.

Load-bearing premise

The result depends on the assumption that summing three axis-wise 1D sweeps preserves the cross-axis spatial interactions a full 3D convolution would capture; if those interactions matter for a pathology, the efficiency gain comes at a representational cost.

Editorial extensions

If this is right

  • The encoder family scales from 0.92M to 15.1M parameters and 34 to 168 GFLOPs, so the same CLIP training and inference pipeline runs on much lighter hardware than CT-ViT's 101.1M parameters and 160.5 GFLOPs.
  • Because cost scales linearly with kernel size, the design makes large-kernel 3D convolutions practical; larger kernels than 7 are a natural next knob to turn.
  • A single linear layer on top of the frozen encoder gives fine-tuned multi-label detection an F1 score of 48.6% for the tiny variant, the best reported in the comparison.
  • The hierarchical four-stage structure preserves multi-scale features that a fixed large-patch tokenizer such as CT-ViT loses, which the authors connect to medical imaging performance.

Reading between the lines

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

  • The effective receptive field of one decomposed block is a 3D cross—each axis is processed independently and the results are summed—so stacked layers must build up oblique or curved structure across multiple blocks; whether that is a strength or weakness for specific pathologies is not tested.
  • The same factorization transfers directly to any 3D modality where local features are roughly axis-aligned, such as MRI, PET, or volumetric video; the paper only evaluates chest CT, so the transfer is an extrapolation.
  • One could push kernel sizes to 31 or 51, as in large-kernel 2D work, at a cost that remains linear, turning the design into a test bed for how much receptive field 3D medical vision-language models actually need.
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

3 major / 8 minor

Summary. DCFormer is a 3D vision encoder that replaces the token mixer in a MetaFormer block with three parallel axis-wise 1D depthwise convolutions along height, width, and depth, whose outputs are normalized and summed (Eq. 7). This decomposition reduces parameters from Ck^3 to 3Ck and FLOPs from 2CHWDk^3 to 6CHWDk. The encoder is hierarchical with four stages, and it is integrated into a CLIP framework with a CXR-BERT text encoder. The paper evaluates this model on the CT-RATE dataset for zero-shot and fine-tuned detection of 18 pathologies, and for image-text retrieval, comparing against CT-ViT, ViT, ConvNeXt, PoolFormer, and TransUNet. The central claim is that the decomposed convolution preserves spatial information while greatly reducing cost, and that DCFormer consistently outperforms the compared encoders.

Significance. The efficiency analysis is correct and clearly presented: the parameter and FLOP formulas for the decomposed convolution are properly derived, and the hierarchical design is well specified. The paper also ships code and uses a unified CLIP training framework across all compared encoders, which is a strength for fair comparison. If the central architectural claim were established, DCFormer would be a useful and practical 3D vision encoder for medical vision-language models. However, the current evidence does not yet establish that the decomposed convolution preserves the spatial information captured by a full 3D convolution, and the reported performance advantages are often small and unsupported by repeated runs or significance testing. The significance is therefore conditional on additional controlled experiments.

major comments (3)
  1. [Formulation of the DCFormer block, Eq. (7)] The central architectural assumption — that X' = X + Norm_h(X*h) + Norm_w(X*w) + Norm_d(X*d) is an adequate replacement for a 3D token mixer — is not tested against the natural control of a full 3D depthwise convolution at a matched parameter/FLOP budget. The effective per-layer operator has support only on the H, W, and D axes, so it cannot represent cross-axis correlations within a single layer. Whether stacking layers and nonlinearities recovers this capacity is an empirical question, and the paper provides no ablation against a standard 3D depthwise convolution (or a full 3D convolution) to support the claim that spatial information is preserved. This is the load-bearing experiment for the core design choice and is currently missing.
  2. [Tables 1 and 3; Abstract and Discussion] All reported results are single runs with no error bars, no multiple seeds, and no statistical significance testing. The margins over strong baselines are often small or reversed. For example, in Table 1 the zero-shot accuracy of DCFormer-tiny (62.0%) is lower than CT-ViT (62.9%) and ConvNeXt-tiny (62.5%), and DCFormer-nano (60.4%) is lower than ConvNeXt-nano (62.2%). In Table 3, TransUNet-naive achieves higher retrieval R@10 (2.36% vs. 2.20%) and R@50 (10.00% vs. 9.54%) than DCFormer-naive. Therefore the claim in the abstract and Discussion that DCFormer 'consistently outperforms' the compared state-of-the-art encoders is not supported by the reported data.
  3. [Image-Text Retrieval (Table 3)] The retrieval protocol is under-specified. It is not stated whether R@k is computed against the full validation set, how many candidates are in the retrieval pool, or whether the reported numbers correspond to the validation split used in the other tasks. Because the retrieval task is one of the three main evaluation pillars, the experimental setup must be described precisely so that the results can be independently reproduced and interpreted.
minor comments (8)
  1. [Eq. (7) and Algorithm 1] The relationship between Eq. (7) and the MetaFormer recurrence in Eq. (2) is unclear: the pseudocode applies BatchNorm3d inside each of the three parallel branches but does not show a normalization applied to X before the three convolutions, while Eq. (7) explicitly normalizes the three convolution outputs. Please clarify whether the block applies a shared normalization to the input of the token mixer and how the per-branch norms relate to Norm1 in Eq. (2).
  2. [Discussion, Med3DVLM validation] The claim that DCFormer has been 'independently validated' in Med3DVLM (ref. 33) is misleading because the author list of Med3DVLM overlaps with the present authors. Please describe this as additional validation by the same group, not an independent evaluation.
  3. [Methods, kernel size notation] The text states that kernel sizes are chosen from {13, 11, 9, 7}, but the architecture table and all reported experiments use kernel size 7 only. Please clarify whether larger kernels were tested or whether the set notation refers to a design choice that was not exercised in the experiments.
  4. [Implementation Details] Please state explicitly whether all baseline encoders (ViT, TransUNet, ConvNeXt, PoolFormer, CT-ViT) were trained with exactly the same training recipe, including the CLIP loss, batch size, learning rate, number of epochs, and absence of scheduling, and whether any per-model hyperparameter tuning was performed.
  5. [Zero-Shot Multi-Abnormality Detection] There is a duplicated word in the sentence 'We compute the cosine similarity between between the CT image embedding and each of the text prompt embeddings.'
  6. [Introduction and Related Work] The novelty claim that DCFormer is 'the first 3D vision encoder to use decomposed convolutions to improve vision-language modeling' should be scoped more carefully and should cite prior work on factorized or axial convolutions, which are closely related to the proposed decomposition.
  7. [Table 3] Table 3 uses the spelling 'naive' while the rest of the text uses 'naïve'; please unify the spelling.
  8. [Discussion, clinical deployability] The Discussion claims that DCFormer 'can run on standard hospital hardware without requiring specialized infrastructure' and is suited for real-time applications, but the paper does not report inference time or latency. Consider adding latency measurements to support this claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the efficiency reduction is arithmetically derived from Eqs. 4-7 and the performance claims are empirical comparisons against external baselines.

full rationale

The central derivation is self-contained. The parameter/FLOP reduction follows by construction from Eqs. 4-7: a k×k×k depthwise kernel with Ck^3 parameters and 2CHWDk^3 FLOPs is replaced by three axis-wise 1D depthwise kernels with 3Ck parameters and 6CHWDk FLOPs; this is an exact arithmetic consequence, not a fitted parameter renamed as a prediction. The zero-shot, fine-tuned, and retrieval results (Tables 1-3) are empirical comparisons against external encoders (ConvNeXt, PoolFormer, TransUNet, ViT, CT-ViT) on CT-RATE, and no target metric is used as an architectural input. The claim that decomposition 'preserves spatial information' is an unverified assumption — no ablation against a full 3D depthwise convolution and no error bars are given — but lack of proof is a correctness/validity concern, not circularity. The only self-citation is Med3DVLM [33] in the Discussion, where the paper calls DCFormer 'independently validated' in a same-author work; that is an overstated independence claim, but it is auxiliary and not load-bearing for the paper's own CT-RATE results. No step in the derivation reduces to its own inputs, so the circularity score is 0.

Assumptions & free parameters 5 free parameters · 3 assumptions · 0 invented entities

The paper's own contribution is the architecture and empirical comparison. It draws on MetaFormer and InceptionNeXt-like design choices, assumes CT-RATE labels are reliable, and fixes the CLIP temperature. No formal proof or external benchmark is offered for the key representational assumption that summed axis-wise convolutions preserve 3D context.

free parameters (5)
  • Base decomposed kernel size k=7 = 7
    Chosen by hand following ConvNeXt and InceptionNeXt; no ablation across k=9, 11, 13 is reported in the experiments despite these being listed as possible.
  • MLP expansion ratio = 4
    Taken from prior transformer and CNN designs; not swept in this paper.
  • Stage depth configurations = [1,1,1,1], [2,2,2,2], [2,3,3,2]
    Hand-designed model family; no scaling study shows these are near-optimal.
  • CLIP temperature tau = 1
    Fixed to 1 in equation (8) instead of learned; affects contrastive alignment and is a choice made by hand.
  • Training recipe = AdamW, lr 1e-5, 15 epochs, no schedule, 8 A100 GPUs
    Training hyperparameters are chosen by hand and not optimized; they influence all reported numbers.
assumptions (3)
  • ad hoc to paper Three parallel axis-wise 1D depthwise convolutions summed additively preserve enough 3D spatial context to serve as a token mixer (Eq. 7).
    This is the core architectural bet; no theoretical justification or ablation against a full 3D depthwise convolution is provided.
  • domain assumption CT-RATE labels and reports are accurate enough to measure zero-shot and fine-tuned pathology detection.
    The dataset labels are extracted from radiology reports by Hamamci et al., and all evaluation inherits that labeling quality.
  • domain assumption CLIP contrastive loss aligns visual and textual embeddings under a fixed temperature tau=1 (Eq. 8).
    Standard CLIP-style training assumption; not independently validated here and the temperature is not learned.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DCFormer: Efficient 3D Vision-Language Modeling with Decomposed Convolutions." pith.science (2026). https://pith.science/paper/LXS3VP5W

@misc{pith2026250205091,
  author       = {Pith},
  title        = {Pith review of: DCFormer: Efficient 3D Vision-Language Modeling with Decomposed Convolutions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LXS3VP5W}},
  note         = {Machine review of arXiv:2502.05091}
}
read the original abstract

Vision-language models (VLMs) have been widely applied to 2D medical image analysis due to their ability to align visual and textual representations. However, extending VLMs to 3D imaging remains computationally challenging. Existing 3D VLMs often rely on Vision Transformers (ViTs), which are computationally expensive due to the quadratic complexity of self-attention, or on 3D convolutions, which require large numbers of parameters and FLOPs as kernel size increases. We introduce DCFormer, an efficient 3D image encoder that factorizes 3D convolutions into three parallel 1D convolutions along the depth, height, and width dimensions. This design preserves spatial information while significantly reducing computational cost. Integrated into a CLIP-based vision-language framework, DCFormer is trained and evaluated on CT-RATE, a dataset of 50,188 paired 3D chest CT volumes and radiology reports. In zero-shot and fine-tuned detection of 18 pathologies, as well as in image-text retrieval tasks, DCFormer consistently outperforms state-of-the-art 3D vision encoders, including CT-ViT, ViT, ConvNeXt, PoolFormer, and TransUNet. These results highlight DCFormer's potential for scalable, clinically deployable 3D medical VLMs. Our code is available at: https://github.com/mirthAI/DCFormer.

Figures

Figures reproduced from arXiv: 2502.05091 by the authors.

Figure 1
Figure 1. Parameter count and computational cost (FLOPs) comparison for 2D and 3D standard and depthwise convolutions across kernel sizes. For simplicity, the number of output channels is fixed at C = 32, and the depth dimension is set to D = 32. are effective at capturing global relationships, their computational complexity scales quadratically with input size, making them inefficient for 3D volumes. Similarly, 3D convolutio… view at source ↗
Figure 2
Figure 2. Parameter count and computational cost (FLOPs) comparison for standard 3D depthwise convolution and decomposed depthwise convolution [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Block illustration of MetaNeXt, ConvNeXt and DCFormer. Metaformer architecture: X ′ = X +TokenMixer(Norm1(X)), (2) X ′′ = X ′ +σ(Norm2(X ′ )W1)W2. (3) Here, Norm1 and Norm2 are typically batch normalization39 or layer normalization40. The TokenMixer serves as the core module for spatial information interaction, W1 and W2 are learnable weights in a two-layer channel MLP, and σ is a non-linear activation41, 42 . To fu… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Hierarchical architecture of DCFormer. the output of the image encoder, while the text encoder output is transformed using a separate linear layer. Finally, visual and textual embeddings are aligned using a contrastive loss function16: LCLIP = − 1 B B ∑ i=1  log exp(…
Figure 5
Figure 5. Figure 5: DCFormer-based CLIP framework: (a) Training with paired CT volumes and reports, (b) Zero-shot inference with text prompts, (c) Fine-tuning for multi-label classification, and (d) Text-to-image retrieval based on embedding similarity [PITH_FULL_IMAGE:figures/full_fig_p…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. GreenRFM: Learning a resource-efficient radiology vision-language foundation model via supervision-centric pre-training

    cs.CV 2026-03 conditional novelty 6.0 of 10

    MUST supervision—LLM-distilled diagnostic labels plus two-stage ubiquitous training—lets a 33M-parameter 3D ResNet-18 reach 84.8 zero-shot AUC on CT-RATE in 24 GPU-hours and transfer across institutions and MRI.

  2. HSENet: Hybrid Spatial Encoding Network for 3D Medical Vision-Language Understanding

    cs.CV 2025-06 conditional novelty 5.0 of 10

    HSENet improves 3D CT vision-language understanding by combining global and local 3D encoders with a centroid-based spatial token compressor, posting state-of-the-art results on CT-RATE and RadGenome-ChestCT.

  3. Taming Vision-Language Models for Medical Image Analysis: A Comprehensive Review

    eess.IV 2025-06 conditional novelty 3.0 of 10

    A survey that classifies vision-language model adaptation for medical imaging into five strategies across eleven tasks, with challenges and future directions.

Reference graph

Works this paper leans on

49 extracted references · 33 canonical work pages · cited by 3 Pith papers

  1. [1]

    Litjens, G. et al. A survey on deep learning in medical image analysis. Med. image analysis 42, 60–88 (2017)

    2017

  2. [2]

    Esteva, A. et al. Deep learning-enabled medical computer vision. NPJ digital medicine 4, 5 (2021). 10/13

    work page 2021

  3. [3]

    & Suk, H.-I

    Shen, D., Wu, G. & Suk, H.-I. Deep learning in medical image analysis. Annu. review biomedical engineering 19, 221–248 (2017)

    2017

  4. [4]

    Fink, O. et al. Potential, challenges and future directions for deep learning in prognostics and health management applications. Eng. Appl. Artif. Intell. 92, 103678 (2020)

    work page 2020

  5. [5]

    Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position

    Fukushima, K. Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position. Biol. cybernetics 36, 193–202 (1980)

    work page 1980

  6. [6]

    Handwritten digit recognition with a back-propagation network

    LeCun, Y .et al. Handwritten digit recognition with a back-propagation network. Adv. neural information processing systems 2 (1989)

    work page 1989

  7. [7]

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

    Dosovitskiy, A. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929 (2020)

    arXiv 2020

  8. [8]

    & Brox, T

    Ronneberger, O., Fischer, P. & Brox, T. U-net: Convolutional networks for biomedical image segmentation. InMedical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part III 18, 234–241 (Springer, 2015)

    work page 2015

Show all 49 references
  1. [9]

    Chen, J. et al. Transunet: Transformers make strong encoders for medical image segmentation. arXiv preprint arXiv:2102.04306 (2021)

    2021 arXiv

  2. [10]

    Cao, H. et al. Swin-unet: Unet-like pure transformer for medical image segmentation. In European conference on computer vision, 205–218 (Springer, 2022)

    2022

  3. [11]

    & Zaiane, O

    Wang, H., Cao, P., Wang, J. & Zaiane, O. R. Uctransnet: rethinking the skip connections in u-net from a channel-wise perspective with transformer. In Proceedings of the AAAI conference on artificial intelligence, vol. 36, 2441–2449 (2022)

    2022

  4. [12]

    C., Mohan, P

    Ates, G. C., Mohan, P. & Celik, E. Dual cross-attention for medical image segmentation. Eng. Appl. Artif. Intell. 126, 107139 (2023)

    2023

  5. [13]

    He, K. et al. Masked autoencoders are scalable vision learners. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 16000–16009 (2022)

    2022

  6. [14]

    Self-supervised pre-training of swin transformers for 3d medical image analysis

    Tang, Y .et al. Self-supervised pre-training of swin transformers for 3d medical image analysis. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 20730–20740 (2022)

    2022

  7. [15]

    Chen, R. J. et al. Scaling vision transformers to gigapixel images via hierarchical self-supervised learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 16144–16155 (2022)

    2022

  8. [16]

    Radford, A. et al. Learning transferable visual models from natural language supervision. In International conference on machine learning, 8748–8763 (PMLR, 2021)

    2021

  9. [17]

    Lin, Z. et al. Frozen clip models are efficient video learners. In European Conference on Computer Vision, 388–404 (Springer, 2022)

    2022

  10. [18]

    & Khan, F

    Thengane, V ., Khan, S., Hayat, M. & Khan, F. Clip model is an efficient continual learner.arXiv preprint arXiv:2210.03114 (2022)

    2022 arXiv

  11. [19]

    & Xiao, Y

    Koleilat, T., Asgariandehkordi, H., Rivaz, H. & Xiao, Y . Medclip-samv2: Towards universal text-driven medical image segmentation. arXiv preprint arXiv:2409.19483 (2024)

    2024 arXiv

  12. [20]

    Y .et al

    Lu, M. Y .et al. A multimodal generative ai copilot for human pathology. Nature 634, 466–473 (2024)

    2024

  13. [21]

    Tiu, E. et al. Expert-level detection of pathologies from unannotated chest x-ray images via self-supervised learning. Nat. Biomed. Eng. 6, 1399–1406 (2022)

    2022

  14. [22]

    Endo, M., Krishnan, R., Krishna, V ., Ng, A. Y . & Rajpurkar, P. Retrieval-based chest x-ray report generation using a pre-trained contrastive language-image model. In Machine Learning for Health, 209–219 (PMLR, 2021)

    2021

  15. [23]

    & Hoogs, A

    Hu, B., Vasu, B. & Hoogs, A. X-mir: Explainable medical image retrieval. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 440–450 (2022)

    2022

  16. [24]

    & De Melo, G

    Eslami, S., Meinel, C. & De Melo, G. Pubmedclip: How much does clip benefit visual question answering in the medical domain? In Findings of the Association for Computational Linguistics: EACL 2023, 1181–1193 (2023)

    2023

  17. [25]

    Huang, Z., Bianchi, F., Yuksekgonul, M., Montine, T. J. & Zou, J. A visual–language foundation model for pathology image analysis using medical twitter. Nat. medicine 29, 2307–2316 (2023)

    2023

  18. [26]

    Hamamci, I. E. et al. A foundation model utilizing chest ct volumes and radiology reports for supervised-level zero-shot detection of abnormalities. CoRR (2024). 11/13

    2024

  19. [27]

    Hamamci, I. E. et al. Generatect: Text-conditional generation of 3d chest ct volumes. InEuropean Conference on Computer Vision, 126–143 (Springer, 2025)

    2025

  20. [28]

    Hatamizadeh, A. et al. Swin unetr: Swin transformers for semantic segmentation of brain tumors in mri images. In International MICCAI brainlesion workshop, 272–284 (Springer, 2021)

    2021

  21. [29]

    Xception: Deep learning with depthwise separable convolutions

    Chollet, F. Xception: Deep learning with depthwise separable convolutions. In Proceedings of the IEEE conference on computer vision and pattern recognition, 1251–1258 (2017)

    2017

  22. [30]

    Boecking, B. et al. Making the most of text semantics to improve biomedical vision–language processing. In European conference on computer vision, 1–21 (Springer, 2022)

    2022

  23. [31]

    Liu, Z. et al. A convnet for the 2020s. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 11976–11986 (2022)

    2022

  24. [32]

    Metaformer is actually what you need for vision

    Yu, W.et al. Metaformer is actually what you need for vision. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 10819–10829 (2022)

    2022

  25. [33]

    C., Gong, K

    Xin, Y ., Ates, G. C., Gong, K. & Shao, W. Med3dvlm: An efficient vision-language model for 3d medical image analysis. arXiv preprint arXiv:2503.20047 (2025)

    2025

  26. [34]

    & Beyer, L

    Zhai, X., Mustafa, B., Kolesnikov, A. & Beyer, L. Sigmoid loss for language image pre-training. In Proceedings of the IEEE/CVF international conference on computer vision, 11975–11986 (2023)

    2023

  27. [35]

    & Lee, Y

    Liu, H., Li, C., Wu, Q. & Lee, Y . J. Visual instruction tuning.Adv. neural information processing systems 36 (2024)

    2024

  28. [36]

    Lai, X. et al. Lisa: Reasoning segmentation via large language model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 9579–9589 (2024)

    2024

  29. [37]

    Metaformer baselines for vision

    Yu, W.et al. Metaformer baselines for vision. IEEE Transactions on Pattern Analysis Mach. Intell.(2023)

    2023

  30. [38]

    & Sun, J

    He, K., Zhang, X., Ren, S. & Sun, J. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, 770–778 (2016)

    2016

  31. [39]

    Batch normalization: Accelerating deep network training by reducing internal covariate shift

    Ioffe, S. Batch normalization: Accelerating deep network training by reducing internal covariate shift. arXiv preprint arXiv:1502.03167 (2015)

    2015 arXiv

  32. [40]

    Ba, J. L. Layer normalization. arXiv preprint arXiv:1607.06450 (2016)

    2016 arXiv

  33. [41]

    & Hinton, G

    Nair, V . & Hinton, G. E. Rectified linear units improve restricted boltzmann machines. In Proceedings of the 27th international conference on machine learning (ICML-10), 807–814 (2010)

    2010

  34. [42]

    & Gimpel, K

    Hendrycks, D. & Gimpel, K. Gaussian error linear units (gelus). arXiv preprint arXiv:1606.08415 (2016)

    2016 arXiv

  35. [43]

    & Wang, X

    Yu, W., Zhou, P., Yan, S. & Wang, X. Inceptionnext: When inception meets convnext. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 5672–5683 (2024)

    2024

  36. [44]

    & Ding, G

    Ding, X., Zhang, X., Han, J. & Ding, G. Scaling up your kernels to 31x31: Revisiting large kernel design in cnns. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 11963–11975 (2022)

    2022

  37. [45]

    Liu, S. et al. More convnets in the 2020s: Scaling up kernels beyond 51x51 using sparsity.arXiv preprint arXiv:2207.03620 (2022)

    2022 arXiv

  38. [46]

    Liu, Z. et al. Swin transformer: Hierarchical vision transformer using shifted windows. In Proceedings of the IEEE/CVF international conference on computer vision, 10012–10022 (2021)

    2021

  39. [47]

    Internimage: Exploring large-scale vision foundation models with deformable convolutions

    Wang, W.et al. Internimage: Exploring large-scale vision foundation models with deformable convolutions. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 14408–14419 (2023)

    2023

  40. [48]

    Dai, Z., Liu, H., Le, Q. V . & Tan, M. Coatnet: Marrying convolution and attention for all data sizes.Adv. neural information processing systems 34, 3965–3977 (2021)

    2021

  41. [49]

    Kingma, D. P. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 (2014). Acknowledgements This work was supported by the Department of Medicine and the Intelligent Clinical Care Center at the University of Florida College of Medicine. The authors expres...

    2014 arXiv

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.