AI Paper Atlas

Research map / Detection and segmentation / CNN features

Scattering Vision Transformer: Spectral Mixing Matters

Badri Narayana Patro, Vijay Srinivas Agneeswaran

NeurIPS 2023 · Detection and segmentation › CNN features · approach: Plain and high-resolution ViTs · 7 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use DTCWT scattering with low-frequency tensor mixing and high-frequency Einstein mixing (ablation)SVT · ImageNet-1KTop-1 accuracy: 83.70 -> 84.20Improved
Place scatter layers first, followed by attention layers (ablation)SVT-H-S · ImageNet-1KTop-1 accuracy: 83.1 -> 84.2Improved
Use initial scatter layers rather than initial convolutional layers (ablation)SVT-H-S · ImageNet-1KTop-1 accuracy: 84.0 -> 84.2Improved
Replace DTCWT scattering with Fourier or wavelet spectral alternatives (ablation)SVT · ImageNet-1KTop-1 accuracy: 83.1; 84.02; 84.09; 83.70 -> 84.20Hurt
Use tensor mixing in both frequency components instead of tensor-low/Einstein-high mixing (ablation)SVT · ImageNet-1KTop-1 accuracy: 83.97 -> 84.20Hurt
Use Einstein mixing in low frequency and tensor mixing in high frequency (ablation)SVT · ImageNet-1KTop-1 accuracy: 83.87 -> 84.20Hurt
Remove high-frequency channel mixing (ablation)SVT_TTEE · ImageNet-1KTop-1 accuracy: 84.06 -> 84.20Hurt
Remove high-frequency token mixing (ablation)SVT_TTEE · ImageNet-1KTop-1 accuracy: 84.12 -> 84.20Hurt
Use all six directional orientations rather than fewer orientations (ablation)SVT-H-S · ImageNet-1KTop-1 accuracy: 83.2; 83.4; 83.7 -> 84.2Improved
Use SVT-H-B instead of DeiT-B/16ImageNet-pretrained transfer learning · CIFAR-100Top-1 accuracy: 90.8 -> 91.2Improved
Use SVT-H-S instead of PVTv2-B2 as the Mask R-CNN backboneMask R-CNN · COCO val2017APᵇ: 45.3 -> 46.0Improved
Use SVT-H-S instead of PVTv2-B2 as the Mask R-CNN backboneMask R-CNN · COCO val2017APᵐ: 41.2 -> 41.9Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy