Research map / Image generation / Super-resolution
QuantSR+: Pushing the Limit of Quantized Image Super-Resolution Networks
Haotong Qin, Xudong Ma, Xianglong Liu, Jie Luo, Jinyang Guo, Michele Magno, Yulun 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 RBD (redistribution-driven bit determination) to DoReFa (ablation) | SRResNet W2A2 · Set5 x4 | PSNR: 30.54 -> 30.75 | Improved |
| add QSA (quantized slimmable architecture) to DoReFa (ablation) | SRResNet W2A2 · Set5 x4 | PSNR: 30.54 -> 31.01 | Improved |
| add SFD (function-localized distillation) to DoReFa (ablation) | SRResNet W2A2 · Set5 x4 | PSNR: 30.54 -> 31.21 | Improved |
| combine RBD and QSA (ablation) | SRResNet W2A2 · Set5 x4 | PSNR: 30.54 -> 31.45 | Improved |
| add SFD on top of RBD+QSA (full QuantSR+) (ablation) | SRResNet W2A2 · Set5 x4 | PSNR: 31.45 -> 31.89 | Improved |
| add RBD to DoReFa (ablation) | SRResNet W2A2 · Set14 x4 | SSIM: 0.7538 -> 0.7514 | Hurt |
| reduce blocks from 32 to 8 with QSA (ablation) | SRResNet W4A4 · Set5 x2 | PSNR: 38.01 -> 37.67 | No clear effect |
| replace QuantSR with QuantSR+ | SwinIR-S W2A2 · Set5 x4 | PSNR: 31.53 -> 31.70 | Improved |
| replace ODM with QuantSR+ | SwinIR-S W2A2 · Urban100 x4 | PSNR: 25.36 -> 25.65 | Improved |
| replace QuantSR with QuantSR+ on diffusion SR | StableSR W4A4 · DIV2K-Val x4 | FID: 88.24 -> 78.62 | Improved |
| add activation shifting and partition from I2NQ (QuantSR+*) | SRResNet W4A4 · Set5 x4 | PSNR: 32.13 -> 32.17 | Improved |
| apply RBD at 1-bit | SRResNet W1A1 · Set5 x4 | PSNR: 32.16 -> 30.32 | Hurt |
Closest papers by meaning
- QuantSR: Accurate Low-bit Quantization for Efficient Image Super-Resolution (NeurIPS 2023 · similarity 0.97)
- SPR$^2$Q: Static Priority-based Rectifier Routing Quantization for Image Super-Resolution (ICLR 2026 · similarity 0.92)
- 2DQuant: Low-bit Post-Training Quantization for Image Super-Resolution (NeurIPS 2024 · similarity 0.91)
- Thinking in Granularity: Dynamic Quantization for Image Super-Resolution by Intriguing Multi-Granularity Clues (AAAI 2025 · similarity 0.91)
- AdaBM: On-the-Fly Adaptive Bit Mapping for Image Super-Resolution (CVPR 2024 · similarity 0.90)
- PassionSR: Post-Training Quantization with Adaptive Scale in One-Step Diffusion based Image Super-Resolution (CVPR 2025 · similarity 0.88)
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.
