Research map / Image generation / Fast diffusion sampling
Invertible Diffusion Models for Compressed Sensing
Bin Chen, Zhenyu Zhang, Weiqi Li, Chen Zhao, Jiwen Yu, Shijie Zhao, Jie Chen, Jian Zhang
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 end-to-end fine-tuning of the diffusion sampling process (ablation) | DDNM with SD v1.5 · Urban100/DIV2K RGB 10% | PSNR (dB): 22.86/25.13 -> 26.17/28.09 | Improved |
| add two-level invertible design (ablation) | IDM* · Urban100/DIV2K RGB 10% | GPU memory (GB): 34.6 -> 13.4 | Improved |
| add two-level invertible design (ablation) | IDM* · Urban100/DIV2K RGB 10% | PSNR (dB): 26.17/28.09 -> 26.21/28.14 | No clear effect |
| add invertibility (ablation) | IDM* · Urban100/DIV2K RGB 10% | training time (h): 411.1 -> 820.7 | Hurt |
| add reuse of pre-trained weights (ablation) | IDM* · Urban100/DIV2K RGB 10% | PSNR (dB): 26.17/28.09 -> 26.57/28.33 | Improved |
| add reuse of pre-trained weights (ablation) | IDM* · Urban100/DIV2K RGB 10% | training time (h): 411.1 -> 12.4 | Improved |
| use SD v1.5 weights instead of unconditional image diffusion model (ablation) | IDM* (invertible, reuse) · Urban100/DIV2K RGB 10% | PSNR (dB): 24.45/26.52 -> 26.52/28.31 | Improved |
| add measurement injectors (ablation) | IDM* · Urban100/DIV2K RGB 10% | PSNR (dB): 26.52/28.31 -> 29.53/30.41 | Improved |
| initialize with back-projection instead of Gaussian noise (ablation) | IDM (pruned) · Urban100/DIV2K RGB 10% | PSNR (dB): 27.21/29.12 -> 29.44/30.37 | Improved |
| back-projection initialization for DDNM variants (ablation) | DDNM (1), (2), (10) · Urban100/DIV2K RGB 10% | PSNR (dB) | No clear effect |
| add network pruning (ablation) | IDM* · Urban100/DIV2K RGB 10% | PSNR (dB): 29.53/30.41 -> 29.44/30.37 | Hurt |
| increase number of sampling steps T beyond 3 (ablation) | IDM · CS PSNR/SSIM | PSNR | No clear effect |
Closest papers by meaning
- Using Powerful Prior Knowledge of Diffusion Model in Deep Unfolding Networks for Image Compressive Sensing (CVPR 2025 · similarity 0.90)
- Adapt and Diffuse: Sample-adaptive Reconstruction via Latent Diffusion Models (ICML 2024 · similarity 0.90)
- OSCAR: One-Step Diffusion Codec Across Multiple Bit-rates (NeurIPS 2025 · similarity 0.89)
- A Diffusion Model with State Estimation for Degradation-Blind Inverse Imaging (AAAI 2024 · similarity 0.89)
- Gradient-free Decoder Inversion in Latent Diffusion Models (NeurIPS 2024 · similarity 0.89)
- Multi-Cross Sampling and Frequency-Division Reconstruction for Image Compressed Sensing (AAAI 2024 · similarity 0.89)
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.
