Research map / Detection and segmentation / CNN features
Activate or Not: Learning Customized Activation
Ningning Ma, Xiangyu Zhang, Ming Liu, Jian Sun
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 |
|---|---|---|---|
| Replace ReLU with meta-ACON | ShuffleNetV2 0.5x · ImageNet 2012 | top-1 error: 39.4 -> 34.8 | Improved |
| Replace ReLU with meta-ACON | ResNet-152 · ImageNet 2012 | top-1 error: 22.3 -> 20.5 | Improved |
| Replace ReLU with ACON-C | ShuffleNetV2 0.5x · ImageNet 2012 | top-1 error: 39.4 -> 37.0 | Improved |
| Use ACON-C instead of ACON-A (Swish) | ResNet-50 · ImageNet 2012 | top-1 error: 23.5 -> 23.2 | Improved |
| Replace ReLU with ACON-A (Swish) | SENet-154 · ImageNet 2012 | top-1 error: 18.95 -> 19.02 | Hurt |
| Replace ReLU with ACON-C | SENet-154 · ImageNet 2012 | top-1 error: 18.95 -> 18.40 | Improved |
| Replace ReLU with meta-ACON | RetinaNet with ResNet-50 · COCO minival | mAP: 35.2 -> 36.5 | Improved |
| Replace ReLU with meta-ACON | PSPNet with ResNet-50 · CityScape | mean_IU: 77.2 -> 78.3 | Improved |
| Generate switching factors channel-wise rather than layer-wise (ablation) | meta-ACON on ShuffleNetV2 0.5x · ImageNet 2012 | top-1 error: 36.3 -> 34.8 | Improved |
| Generate switching factors pixel-wise rather than layer-wise (ablation) | meta-ACON on ShuffleNetV2 0.5x · ImageNet 2012 | top-1 error: 36.3 -> 37.2 | No clear effect |
Closest papers by meaning
- Active Convolution: Learning the Shape of Convolution for Image Classification (CVPR 2017 · similarity 0.85)
- SAU: Smooth Activation Function Using Convolution with Approximate Identities (ECCV 2022 · similarity 0.85)
- Smooth Maximum Unit: Smooth Activation Function for Deep Networks Using Smoothing Maximum Technique (CVPR 2022 · similarity 0.84)
- L-SWAG: Layer-Sample Wise Activation with Gradients Information for Zero-Shot NAS on Vision Transformers (CVPR 2025 · similarity 0.83)
- Inducing and Exploiting Activation Sparsity for Fast Inference on Deep Neural Networks (ICML 2020 · similarity 0.83)
- Do We Always Need the Simplicity Bias? Looking for Optimal Inductive Biases in the Wild (CVPR 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.
