Research map / Architectures and efficiency / CNN design
Learned Image Compression With Mixed Transformer-CNN Architectures
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 |
|---|---|---|---|
| combine CNN residual blocks and Swin Transformer blocks in parallel using TCM (ablation) | medium LIC model · Kodak | rate-distortion performance | Helped |
| add SWAtten to the channel-wise entropy model (ablation) | LIC model · Kodak | rate-distortion performance | Helped |
| add SWAtten (ablation) | LIC model · Kodak kodim01 | scaled mean absolute pixel deviation: 0.451 -> 0.389 | Helped |
| add SWAtten (ablation) | LIC model · Kodak kodim05 | scaled mean absolute pixel deviation: 0.422 -> 0.365 | Helped |
| add channel squeeze/unsqueeze to SWAtten (ablation) | Ours (Large) · Kodak | parameters (M): 160.45 -> 75.89 | Helped |
| add channel squeeze/unsqueeze to SWAtten (ablation) | Ours (Large) · Kodak | BD-rate: -12.54 -> -12.30 | No clear effect |
| reduce channel-wise entropy-model slices from 10 to 5 (ablation) | LIC entropy model with SWAtten · Kodak | rate-distortion performance | Mixed |
| add CNN attention-map generation to SWAtten (ablation) | SWAtten · Kodak | PSNR: about 0.1dB | Helped |
| add the Swin Transformer to SWAtten (ablation) | SWAtten · Kodak | PSNR: about 0.1dB | Helped |
| use SWAtten instead of NonlocalAtten, LocalAtten, or WAtten (ablation) | channel-wise entropy model · Kodak | rate-distortion performance | Helped |
| use TCM in the main path (ablation) | LIC model with spatial-wise entropy model · Kodak | rate-distortion performance | Helped |
Closest papers by meaning
- Few-Shot Domain Adaptation for Learned Image Compression (AAAI 2025 · similarity 0.88)
- Frequency-Aware Transformer for Learned Image Compression (ICLR 2024 · similarity 0.88)
- Causal Context Adjustment Loss for Learned Image Compression (NeurIPS 2024 · similarity 0.88)
- Linear Attention Modeling for Learned Image Compression (CVPR 2025 · similarity 0.88)
- Balanced Rate-Distortion Optimization in Learned Image Compression (CVPR 2025 · similarity 0.88)
- On Disentangled Training for Nonlinear Transform in Learned Image Compression (ICLR 2025 · similarity 0.87)
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.