Research map / Detection and segmentation / CNN features
Gated Channel Transformation for Visual Recognition
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 GCT before each convolutional layer | ResNet-50 · ImageNet 2012 val | top-1 error (%): 23.8 -> 22.7 | Helped |
| Add GCT before each convolutional layer | ResNet-50 · COCO eval2017 (minival), Mask R-CNN | box AP: 37.8 -> 39.8 | Helped |
| Add GCT before each convolutional layer | ResNet-50 · COCO eval2017 (minival), Mask R-CNN | mask AP: 34.2 -> 36.0 | Helped |
| Add GCT at the last two convolutional layers in each Res-Block | ResNet-50 (3D) · Kinetics-400 | top-1 accuracy (%): 74.6 -> 75.1 | Helped |
| Replace the L2 embedding norm with L-infinity (ablation) | ResNet-50 + GCT · ImageNet 2012 val | top-1 error (%): 22.7 -> 23.1 | Hurt |
| Replace the L2 embedding norm with L1 (ablation) | ResNet-50 + GCT · ImageNet 2012 val | top-1 error (%): 22.7 -> 22.8 | Hurt |
| Replace channel L2 normalization with mean and variance normalization (ablation) | ResNet-50 + GCT · ImageNet 2012 val | top-1 error (%): 22.7 -> 23.7 | Hurt |
| Replace channel L2 normalization with L1 normalization (ablation) | ResNet-50 + GCT · ImageNet 2012 val | top-1 error (%): 22.7 -> 22.9 | Hurt |
| Replace 1+tanh gate adaptation with sigmoid (ablation) | ResNet-50 + GCT · ImageNet 2012 val | top-1 error (%): 22.7 -> 22.9 | Hurt |
| Replace 1+tanh gate adaptation with 1+ELU (ablation) | ResNet-50 + GCT · ImageNet 2012 val | top-1 error (%): 22.7 -> 22.7 | No clear effect |
| Place GCT after BN instead of before convolution (ablation) | ResNet-50 + GCT · ImageNet 2012 val | top-1 error (%): 22.7 -> 23.1 | Hurt |
| Place GCT before BN instead of before convolution (ablation) | ResNet-50 + GCT · ImageNet 2012 val | top-1 error (%): 22.7 -> 23.1 | Hurt |
Closest papers by meaning
- Squeeze-and-Excitation Networks (CVPR 2018 · similarity 0.88)
- Channel Importance Matters in Few-Shot Image Classification (ICML 2022 · similarity 0.87)
- Visual Transformer with Differentiable Channel Selection: An Information Bottleneck Inspired Approach (ICML 2024 · similarity 0.84)
- Channel Vision Transformers: An Image Is Worth 1 x 16 x 16 Words (ICLR 2024 · similarity 0.84)
- ChAda-ViT : Channel Adaptive Attention for Joint Representation Learning of Heterogeneous Microscopy Images (CVPR 2024 · similarity 0.84)
- AsCAN: Asymmetric Convolution-Attention Networks for Efficient Recognition and Generation (NeurIPS 2024 · similarity 0.84)
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.