Research map / Detection and segmentation / CNN features
Volumetric Transformer Networks
Seungryong Kim, Sabine Ssstrunk, Mathieu Salzmann
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add channel squeeze and expansion modules (ablation) | VTNs (ResNet-50) · CUB-Birds | accuracy: 80.1 -> 83.1 | Improved |
| add group sampling and normalization module (ablation) | VTNs (ResNet-50) · CUB-Birds | accuracy: 77.2 -> 83.1 | Improved |
| add group sampling and normalization module (ablation) | VTNs (ResNet-50) · FGVC-Aircraft | accuracy: 82.4 -> 89.2 | Improved |
| add transformation probability inference module (ablation) | VTNs (ResNet-50) · CUB-Birds | accuracy: 79.0 -> 83.1 | Improved |
| add consistency loss L_cons (ablation) | VTNs (ResNet-50) · CUB-Birds | accuracy: 82.4 -> 83.1 | Improved |
| add consistency loss L_cons (ablation) | VTNs (ResNet-50) · FGVC-Aircraft | accuracy: 84.9 -> 89.2 | Improved |
| add consistency loss L_cons (ablation) | VTNs (ResNet-50) · Stanford-Cars | accuracy: 82.1 -> 82.7 | No clear effect |
| add consistency loss L_cons (ablation) | VTNs (VGG-19) · CUB-Birds | accuracy: 79.2 -> 80.4 | Improved |
| replace single shared warping field (STNs) with channel-wise warping (VTNs) | ResNet-50 · CUB-Birds | accuracy: 76.5 -> 83.1 | Improved |
| add VTNs to iSQRT-COV pooling | iSQRT-COV (ResNet-50) · CUB-Birds | accuracy: 88.1 -> 89.6 | Improved |
| add VTNs to iSQRT-COV pooling | iSQRT-COV (ResNet-50) · Stanford-Cars | accuracy: 90.0 -> 93.3 | Improved |
| add VTNs to TASN features on iNaturalist-2017 | TASN · iNaturalist-2017 | accuracy: 66.2 -> 68.2 | Improved |
Closest papers by meaning
- Visual Transformers: Where Do Transformers Really Belong in Vision Models? (ICCV 2021 · similarity 0.87)
- TI-Pooling: Transformation-Invariant Pooling for Feature Learning in Convolutional Neural Networks (CVPR 2016 · similarity 0.86)
- Polar Transformer Networks (ICLR 2018 · similarity 0.86)
- Understanding Geometry of Encoder-Decoder CNNs (ICML 2019 · similarity 0.86)
- Warped Convolutions: Efficient Invariance to Spatial Transformations (ICML 2017 · similarity 0.86)
- Exploiting Invariance in Training Deep Neural Networks (AAAI 2022 · similarity 0.86)
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.
