Research map / Architectures and efficiency / CNN design
Identity Mappings in Deep Residual Networks
Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun
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 full pre-activation (BN-ReLU before weights) (ablation) | ResNet-110 · CIFAR-10 | error (%): 6.61 -> 6.37 | Improved |
| add full pre-activation (ablation) | ResNet-164 · CIFAR-10 | error (%): 5.93 -> 5.46 | Improved |
| add full pre-activation | ResNet-1001 · CIFAR-10 | error (%): 7.61 -> 4.92 | Improved |
| add full pre-activation | ResNet-1001 · CIFAR-100 | error (%): 27.82 -> 22.71 | Improved |
| add full pre-activation | ResNet-200 · ImageNet val | top-1 error (%): 21.8 -> 20.7 | Improved |
| add full pre-activation | ResNet-152 · ImageNet val | top-1 error (%): 21.3 -> 21.1 | Improved |
| move BN after addition (ablation) | ResNet-110 · CIFAR-10 | error (%): 6.61 -> 8.17 | Hurt |
| move ReLU before addition (ablation) | ResNet-110 · CIFAR-10 | error (%): 6.61 -> 7.84 | Hurt |
| use ReLU-only pre-activation instead of full pre-activation (ablation) | ResNet-164 · CIFAR-10 | error (%): 5.46 -> 5.91 | Hurt |
| scale shortcut and F by 0.5 (frozen gating) (ablation) | ResNet-110 · CIFAR-10 | error (%): 6.61 -> 12.35 | Hurt |
| replace identity shortcut with 1x1 conv (ablation) | ResNet-110 · CIFAR-10 | error (%): 6.61 -> 12.22 | Hurt |
| use exclusive gating with b_g=-6 on shortcuts (ablation) | ResNet-110 · CIFAR-10 | error (%): 6.61 -> 8.70 | Hurt |
Closest papers by meaning
- Learning Strict Identity Mappings in Deep Residual Networks (CVPR 2018 · similarity 0.89)
- Identity Matters in Deep Learning (ICLR 2017 · similarity 0.87)
- Norm-Preservation: Why Residual Networks Can Become Extremely Deep? (TPAMI 2020 · similarity 0.86)
- Residual Connections Encourage Iterative Inference (ICLR 2018 · similarity 0.86)
- Residual Alignment: Uncovering the Mechanisms of Residual Networks (NeurIPS 2023 · similarity 0.86)
- Revisiting Residual Connections: Orthogonal Updates for Stable and Efficient Deep Networks (NeurIPS 2025 · similarity 0.83)
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.
