REVIEW 4 major objections 6 minor 2 cited by
SRE-Conv: Symmetric Rotation Equivariant Convolution for Biomedical Image Classification
T0 review · 4 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash
Pith's one-line read Radial-band convolution wins all 16 rotated medical-image benchmarks.
desk verdict Solid engineering extension of a known kernel idea, with a broad benchmark, but the kernel-size confound and overclaimed exactness leave the mechanism unproven. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The central object is the SRE-Conv kernel, defined as $K = M_I \Theta$, where $\Theta \in \mathbb{R}^{[C,b]}$ contains the trainable weights and $M_I \in \mathbb{R}^{[b,k,k]}$ is a fixed binary index matrix that assigns each kernel position to one of $b = \lfloor k/2 \rfloor + 2$ radial bands by Euclidean distance from the center, with the four corners zeroed to create a circular support. This construction reduces trainable parameters from $O(C k^2)$ to $O(C k)$, allows arbitrary kernel sizes, and preserves the same floating-point operations as standard convolution because the full kernel is precomputed at inference. The equivariance claim rests on the radial symmetry of the bands combined with convolution's translation equivariance.
What would settle it
Take a trained SRE-Conv layer, rotate a fixed input image by an angle that is not a multiple of 90 degrees, run both versions through the layer, rotate the second feature map back, and measure the maximum absolute difference: exact equivariance predicts a difference at machine precision, while interpolation and band discretization would produce a nonzero gap. A second check is to compare the kernel $K$ itself under an arbitrary rotation: if the rotated kernel differs from the original at any non-90-degree angle, the claimed exact equivariance in Section 2.1 is falsified at that angle.
Extended reading notes
Core claim
The paper's central claim is that a convolution kernel parameterized by $b$ non-overlapping circular bands, defined as equal-width ranges of Euclidean distance from the kernel center, produces feature maps that stay consistent when the input is rotated or reflected. Because ordinary convolution is translation-equivariant, sliding this radially symmetric kernel over the image extends the local symmetry to global rotation and reflection equivariance, and global adaptive pooling makes the final classification invariant. Across all 16 MedMNISTv2 tasks, the resulting SRE-CNN improves rotated-test accuracy over both conventional ResNet baselines and five rotation-equivariant competitors in 2D, and over two 3D baselines, with no statistically significant drop on original test sets overall. The paper acknowledges that on datasets with a strong anatomical orientation prior, such as OrganA in 2D and Organ in 3D, the symmetric constraint can lower original-test accuracy.
Load-bearing premise
The load-bearing premise is that grouping kernel positions into radial distance bands makes the convolution exactly rotation equivariant; on a discrete pixel grid this is exact only for rotations that map grid points onto grid points, and the test rotations (10 degrees in 2D, 30 degrees about each axis in 3D) require interpolation, so the claimed mechanism is approximate at those angles.
Editorial extensions
If this is right
- Rotation and reflection data augmentation may become unnecessary for biomedical classifiers built on SRE-Conv, since rotated and reflected test performance matches original performance.
- Large kernels become affordable: because parameters scale linearly with kernel size rather than quadratically in 2D or cubically in 3D, SRE-Conv can use 9x9 or larger kernels to enlarge receptive fields without blowing up model size.
- The same construction extends to 3D, where parameter savings are even larger, making it practical for volumetric medical images with arbitrary orientation.
- SRE-Conv can replace standard convolution layers in existing CNN backbones, so the accuracy and efficiency trade-off is available without architectural redesign.
- On datasets with a strong anatomical orientation prior, the symmetric constraint can hurt original-test accuracy, so pure SRE-CNNs are best suited to orientation-agnostic tasks or need additional orientation-aware components.
Reading between the lines
- Because the equivariance is exact only for rotations that map grid points to grid points, such as multiples of 90 degrees, the strong gains on 10-degree and 30-degree rotated inputs are partly a discrete approximation; a direct pixel-level invariance test on non-grid rotations would clarify how much comes from exact symmetry versus learned robustness.
- The band-binning parameterization is effectively a radial basis with hard boundaries, so replacing hard bands with smooth radial basis functions could preserve the parameter savings while reducing interpolation artifacts at arbitrary angles.
- A hybrid that combines SRE-Conv with a small orientation-estimating branch could recover the lost performance on orientation-prior datasets like OrganA while keeping rotation robustness on orientation-free tasks.
- The same kernel construction could be applied beyond classification, such as to segmentation or detection backbones, where local rotation equivariance may help when individual objects rotate within a larger scene.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper introduces SRE-Conv, a convolution kernel parameterized by radial distance bands so that all weights within a symmetric band are shared. The authors construct fully convolutional networks using SRE-Conv and evaluate them on 16 MedMNISTv2 tasks (10 2D, 6 3D), reporting improved classification accuracy on rotated test sets for all 16 tasks compared with ResNet-based and equivariant baselines, while using fewer parameters and less GPU memory. The paper also shows qualitative feature-map alignment under rotation and makes the code publicly available.
Significance. If the empirical claims are established, the method would offer a simple and lightweight way to build rotation robustness into CNN backbones for biomedical imaging, and the public code is a useful contribution. The parameter reduction from weight sharing (O(C·k) instead of O(C·k²)) is real, and the authors are careful to avoid geometric data augmentation, making the comparisons directly relevant to equivariant learning. However, the central evidence is weakened by a kernel-size confound between the SRE models (9×9 and 5×5 kernels) and the baselines (3×3 kernels), and by the paper's overstated claim of exact equivariance for arbitrary rotation angles on a discrete grid. The lack of any variance information further tempers confidence in the reported gains.
major comments (4)
- [§3.1 and §3.2, Tables 1–2] The comparison is confounded by kernel size. SRE-R18 and SRE-R3D-18 use kernel sizes [9,9,5,5] for all stages, while the R18/R3D-18 baselines use 3×3 kernels (7×7 in the first conv). The rotated-accuracy advantage could therefore arise from the larger receptive field rather than from the radial band-sharing mechanism. Please add ablations with kernel-size-matched models: (i) SRE-Conv with 3×3 kernels (or the baseline's kernel sizes) and (ii) standard convolution with [9,9,5,5] kernels. Also report FLOPs and inference time for all models; as it stands, the statement in §2.2 that SRE-Conv has 'the same number of floating operations (FLOPs) as traditional convolution' is true only relative to a standard kernel of the same size, not relative to the 3×3 baselines actually compared.
- [§2.1] The claim of exact rotational equivariance is not valid for the non-lattice rotations used in the evaluation. On a discrete pixel grid, a kernel constructed by binning Euclidean distances is invariant to rotations that map grid points to grid points (multiples of 90°), but the rotated test sets use 10° increments (2D) and 30° increments about each axis (3D), which require interpolation. The band-binning makes the kernel a radial step function, so the equivariance is approximate for these angles. Please state this qualification explicitly, provide a quantitative characterization of the approximation (e.g., how rotated accuracy varies with rotation angle), and avoid the unqualified statement that the arrangement 'ensures local equivariance'.
- [§3.2 and Tables 1–2] No per-run variance is reported. All accuracy numbers appear to come from a single training run per model and dataset, and the paired t-tests are computed across the 10 (or 6) datasets rather than across repeated runs. Thus the reported p-values only show that the mean differences across datasets are unlikely to be zero under a paired test; they do not account for training stochasticity. Please report results from multiple seeds with standard deviations or confidence intervals, or clearly justify the single-run protocol.
- [§2.2 and Abstract] The efficiency claim is overstated. While the parameter count is indeed reduced to O(C·k), the actual convolution is a dense operation on a k×k kernel, so a 9×9 SRE-Conv performs roughly 9× the multiply-adds of a 3×3 standard convolution at the same resolution. The paper reports lower GPU memory for SRE-R18, but this does not translate to lower compute. The abstract's phrase 'increasing efficiency' should be qualified as parameter and memory efficiency, and the trade-off with computational cost should be disclosed in the experimental comparison.
minor comments (6)
- [Abstract] The phrase 'increasing efficiency with fewer parameters and reduced memory footprint' leaves 'efficiency' ambiguous; consider using 'parameter efficiency' and explicitly stating that computational cost is comparable to a same-size standard convolution.
- [§2.1] The notation involving C is confusing: 'we denote C for Cin and Cout for simplicity' is followed by K ∈ R[C,k,k], which does not represent the usual [Cin,Cout,k,k] shape or a shared kernel unless Cin=Cout. Please clarify the tensor notation.
- [§2.1] The construction of the binary index matrix is described as 'assigning 1 to each row-column index that has equal distance values in D'; this is ambiguous because D contains many distinct distance values. Please specify the thresholding or band-assignment rule precisely.
- [§3.1] The text states that OrganC and OrganS are skipped because they are repeated with OrganA and Organ3D, but OrganS is not a duplicate of OrganA in MedMNISTv2; please clarify the relationship between the skipped and included datasets.
- [Figure 2] The description of unrotating feature maps should state the interpolation method used for the 60° increments, since these are not lattice rotations and the qualitative alignment depends on interpolation.
- [Tables 1–2] The table layout for memory usage is inconsistent: in Table 1 the 'Mem.' column appears only in the second block, and it is unclear whether 'Mem.' applies to each model row or only to the listed row. Please format the tables so that every model has a memory value.
Circularity Check
No significant circularity: the rotation invariance is built into the kernel by design, and the accuracy claims are external benchmark results.
full rationale
The paper's rotation equivariance claim follows directly from the construction of the SRE-Conv kernel: the kernel is defined with distance-based circular bands (K = M_I Θ), so it is centrally symmetric by construction. This is a design property, not a hidden prediction, and the paper does not pretend that equivariance is learned or independently derived. The central empirical claim—improved accuracy on rotated test sets across 16 MedMNISTv2 tasks—is an external benchmark result. It is not produced by fitting a parameter to the test set, nor is it equivalent to the construction by definition: a symmetric kernel can still fail on rotated data if it is not discriminative, and indeed the paper reports degraded original-test performance on orientation-sensitive datasets like OrganA. No load-bearing self-citations or imported uniqueness theorems appear; the cited prior works ([11], [12], [13]) are background, not circular justification. The larger kernel sizes used for SRE-Conv ([9,9,5,5]) versus 3x3 baselines are a potential experimental confound for isolating the source of accuracy gain, but that is a validity concern, not a circularity in the derivation chain.
Assumptions & free parameters
free parameters (2)
- kernel sizes per stage =
[9,9,5,5] for 2D and 3D
- number of bands b =
floor(k/2)+2
assumptions (4)
- standard math Translation equivariance of convolution
- domain assumption A circularly symmetric kernel is rotation equivariant
- domain assumption Stride-1 convolutions plus equivariant pooling preserve equivariance across layers
- standard math Global average pooling produces rotation-invariant classification features
Cite this review
Pith. "Pith review of SRE-Conv: Symmetric Rotation Equivariant Convolution for Biomedical Image Classification." pith.science (2026). https://pith.science/paper/VS2DCLT5
@misc{pith2026250109753,
author = {Pith},
title = {Pith review of: SRE-Conv: Symmetric Rotation Equivariant Convolution for Biomedical Image Classification},
year = {2026},
howpublished = {\url{https://pith.science/paper/VS2DCLT5}},
note = {Machine review of arXiv:2501.09753}
}
read the original abstract
Convolutional neural networks (CNNs) are essential tools for computer vision tasks, but they lack traditionally desired properties of extracted features that could further improve model performance, e.g., rotational equivariance. Such properties are ubiquitous in biomedical images, which often lack explicit orientation. While current work largely relies on data augmentation or explicit modules to capture orientation information, this comes at the expense of increased training costs or ineffective approximations of the desired equivariance. To overcome these challenges, we propose a novel and efficient implementation of the Symmetric Rotation-Equivariant (SRE) Convolution (SRE-Conv) kernel, designed to learn rotation-invariant features while simultaneously compressing the model size. The SRE-Conv kernel can easily be incorporated into any CNN backbone. We validate the ability of a deep SRE-CNN to capture equivariance to rotation using the public MedMNISTv2 dataset (16 total tasks). SRE-Conv-CNN demonstrated improved rotated image classification performance accuracy on all 16 test datasets in both 2D and 3D images, all while increasing efficiency with fewer parameters and reduced memory footprint. The code is available at https://github.com/XYPB/SRE-Conv.
Forward citations
Cited by 2 Pith papers
-
Improved Vessel Segmentation with Symmetric Rotation-Equivariant U-Net
An equivariant U-Net using symmetric rotation-equivariant kernels improves retinal vessel segmentation on rotated images while using far fewer parameters than baseline networks.
-
Equivariant Imaging Biomarkers for Robust Unsupervised Segmentation of Histopathology
A rotation-equivariant convolutional network produces more rotation-consistent unsupervised segmentations of prostate histopathology images than standard or group-equivariant baselines.
Reference graph
Works this paper leans on
-
[11]
Group equivariant con- volutional networks,
Taco Cohen and Max Welling, “Group equivariant con- volutional networks,” in International conference on machine learning. PMLR, 2016, pp. 2990–2999
2016
-
[12]
Yanzhao Zhou, Qixiang Ye, Qiang Qiu, and Jianbin Jiao, “Oriented response networks,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 519–528
work page 2017
-
[1]
INTRODUCTION Convolution layers play a pivotal role in the successful ap- plication of deep learning methods to computer vision and biomedical image analysis tasks, where the learned convolu- tional kernel weights demonstrate frequency- and orientation- specific responses to images [1]. The significance of convo- lution kernels lies in capturing local pat...
arXiv 2025
-
[2]
METHODS 2.1. Symmetric Rotation-Equivariant (SRE) Convolution We employ a centrally symmetric kernel [13] for equivari- ance (Fig. 1). The SRE-Conv kernel is parameterized using non-overlapping discrete circular bands based on their Eu- clidean distance from the kernel center, where each band rep- resents one trainable parameter. This arrangement ensures ...
-
[3]
EXPERIMENTS AND RESULTS 3.1. Experimental Setup Datasets: To evaluate the effectiveness of our model across various medical imaging modalities, we validate SRE-CNN on the public MedMNISTv2 [15] classification dataset. MedMNISTv2 comprises 12 2D and 6 3D medical imaging datasets, covering a wide spectrum of clinical applications. Benchmarking on such a div...
-
[4]
DISCUSSION AND CONCLUSION By incorporating rotational equivariance directly into the convolutional kernel design, deep learning models can learn equivariant features that are invariant to rotation and reflec- tion, enhancing their robustness to real-world scenarios where objects may vary in orientation. Our results demonstrate that our SRE-Conv layers wit...
-
[5]
Imagenet classification with deep convolutional neural networks,
Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hin- ton, “Imagenet classification with deep convolutional neural networks,” Advances in neural information pro- cessing systems, vol. 25, 2012
2012
-
[6]
Revisiting data augmenta- tion for rotational invariance in convolutional neural net- works,
Facundo Quiroga, Franco Ronchetti, Laura Lanzarini, and Aurelio F Bariviera, “Revisiting data augmenta- tion for rotational invariance in convolutional neural net- works,” in Modelling and Simulation in Management Sciences: Proceedings of the International Conference on Modelling and Simulation in Management Sciences (MS-18). Springer, 2020, pp. 127–141
work page 2020
Show all 21 references
-
[7]
Spatial transformer networks,
Max Jaderberg, Karen Simonyan, Andrew Zisserman, et al., “Spatial transformer networks,” Advances in neu- ral information processing systems, vol. 28, 2015
2015
-
[8]
Gradient-aligned convolution neural network,
You Hao, Ping Hu, Shirui Li, Jayaram K Udupa, Yubing Tong, and Hua Li, “Gradient-aligned convolution neural network,” Pattern Recognition, vol. 122, pp. 108354, 2022
2022
-
[9]
Harmonic net- works: Deep translation and rotation equivariance,
Daniel E Worrall, Stephan J Garbin, Daniyar Tur- mukhambetov, and Gabriel J Brostow, “Harmonic net- works: Deep translation and rotation equivariance,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 5028–5037
2017
-
[10]
General e (2)- equivariant steerable cnns,
Maurice Weiler and Gabriele Cesa, “General e (2)- equivariant steerable cnns,” Advances in neural infor- mation processing systems, vol. 32, 2019
2019
-
[13]
RIC-CNN: Rotation- Invariant coordinate convolutional neural network,
Hanlin Mo and Guoying Zhao, “RIC-CNN: Rotation- Invariant coordinate convolutional neural network,”Pat- tern Recognit., vol. 146, pp. 109994, Feb. 2024
2024
-
[14]
Spin: Simplify- ing polar invariance for neural networks application to vision-based irradiance forecasting,
Quentin Paletta, Anthony Hu, Guillaume Arbod, Philippe Blanc, and Joan Lasenby, “Spin: Simplify- ing polar invariance for neural networks application to vision-based irradiance forecasting,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition...
2022
-
[15]
Use of sym- metric kernels for convolutional neural networks,
Viacheslav Dudar and Vladimir Semenov, “Use of sym- metric kernels for convolutional neural networks,” in Recent Developments in Data Science and Intelligent Analysis of Information: Proceedings of the XVIII In- ternational Conference on Data Science and Intelligent Analysis o...
2018
-
[16]
Rotated ring, radial and depth wise separable radial convolutions,
Wolfgang Fuhl and Enkelejda Kasneci, “Rotated ring, radial and depth wise separable radial convolutions,” in 2021 International Joint Conference on Neural Net- works (IJCNN). IEEE, 2021, pp. 1–8
2021
-
[17]
Theory of edge detection,
D Marr and E Hildreth, “Theory of edge detection,” Proc. R. Soc. Lond. B Biol. Sci., vol. 207, no. 1167, pp. 187–217, Feb. 1980
1980
-
[18]
Understanding the effective receptive field in deep convolutional neural networks,
Wenjie Luo, Yujia Li, Raquel Urtasun, and Richard Zemel, “Understanding the effective receptive field in deep convolutional neural networks,” Proceedings of the 30th International Conference on Neural Informa- tion Processing Systems, pp. 4898–4906, Jan. 2017
2017
-
[19]
Medmnist v2-a large-scale lightweight benchmark for 2d and 3d biomedical image classification,
Jiancheng Yang, Rui Shi, Donglai Wei, Zequan Liu, Lin Zhao, Bilian Ke, Hanspeter Pfister, and Bingbing Ni, “Medmnist v2-a large-scale lightweight benchmark for 2d and 3d biomedical image classification,” Scientific Data, vol. 10, no. 1, pp. 41, 2023
2023
-
[20]
Deep residual learning for image recognition,
Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun, “Deep residual learning for image recognition,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 770–778
2016
-
[21]
A closer look at spatiotemporal convolutions for action recognition,
Du Tran, Heng Wang, Lorenzo Torresani, Jamie Ray, Yann LeCun, and Manohar Paluri, “A closer look at spatiotemporal convolutions for action recognition,” in Proceedings of the IEEE conference on Computer Vision and Pattern Recognition, 2018, pp. 6450–6459
2018
Reviewed August 10, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.