Research map / Architectures and efficiency / CNN design
Learning a Wavelet-Like Auto-Encoder to Accelerate Deep Neural Networks
Read the paperCheck 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 |
|---|---|---|---|
| Replace full-resolution VGG16 input with WAE's two half-resolution channels and lightweight fusion | VGG16-Net · ImageNet val | top-5 error (%): 11.65 -> 11.87 | Hurt |
| Replace full-resolution VGG16 input with WAE's two half-resolution channels and lightweight fusion | VGG16-Net · ImageNet val | CPU speed-up rate: 3.13× | Helped |
| Use learned WAE decomposition rather than fixed DWT channels | VGG16-Net · ImageNet val | top-5 error (%): 14.42 -> 11.87 | Helped |
| Add transform reconstruction and high-frequency energy constraints | WAE classification · ImageNet val | top-5 error (%): 12.98 -> 11.87 | Helped |
| Add joint fine tuning of WAE and classifier (ablation) | WAE with VGG16-Net · ImageNet val | top-5 error (%): 12.21 -> 11.87 | Helped |
| Fuse high-frequency channel I_H with low-frequency I_L instead of using I_L alone (ablation) | WAE classifier · ImageNet val | top-5 error: 14.20 -> 11.87 | Helped |
| Use learned I_L instead of resized image I_R | VGG16-Net · ImageNet val | top-5 error: 15.73 -> 14.20 | Helped |
| Use learned I_L instead of DWT approximation cA | VGG16-Net · ImageNet val | top-5 error: 15.92 -> 14.20 | Helped |
| Apply WAE decomposition rather than standard VGG16 | VGG16-Net · CACD test | accuracy (%): 95.91 -> 96.13 | Helped |
| Apply WAE decomposition rather than standard VGG16 to noisy images | VGG16-Net · CACD test, V=0.05 | accuracy (%): 45.10 -> 55.91 | Helped |
| Replace full-resolution ResNet-50 input with WAE decomposition | ResNet-50 · ImageNet val | top-5 error (%): 8.86 -> 9.66 | Hurt |
Closest papers by meaning
- Deep Decoder: Concise Image Representations from Untrained Non-convolutional Networks (ICLR 2019 · similarity 0.85)
- Wavelet Integrated CNNs for Noise-Robust Image Classification (CVPR 2020 · similarity 0.85)
- Multiscale Low-Frequency Memory Network for Improved Feature Extraction in Convolutional Neural Networks (AAAI 2024 · similarity 0.84)
- Auto-Compressing Networks (NeurIPS 2025 · similarity 0.84)
- Task-Driven Wavelets using Constrained Empirical Risk Minimization (CVPR 2024 · similarity 0.84)
- Extending Video Masked Autoencoders to 128 frames (NeurIPS 2024 · 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.