Research map / Image generation / Super-resolution
HarmoQ: Harmonized Post-Training Quantization for High-Fidelity Image Super-Resolution
Hongjun Wang, Jiyuan Chen, Xuan Song, Yinqiang Zheng
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 structural residual calibration (SRC) to HSO+ABR (ablation) | HarmoQ · Set5 x2 (SwinIR-light) | PSNR: 37.01 -> 37.98 | Improved |
| add harmonized scale optimization (HSO) to SRC+ABR (ablation) | HarmoQ · Set5 x2 (SwinIR-light) | PSNR: 36.78 -> 37.98 | Improved |
| add adaptive boundary refinement (ABR) to SRC+HSO (ablation) | HarmoQ · Set5 x2 (SwinIR-light) | PSNR: 37.12 -> 37.98 | Improved |
| add ABR alone to baseline SwinIR-Light (ablation) | HarmoQ · Set5 x2 (SwinIR-light) | PSNR: 35.12 -> 35.89 | Improved |
| add HSO alone vs SRC alone (ablation) | HarmoQ · Set5 x2 (SwinIR-light) | PSNR: SRC 36.24 vs HSO 36.45 | Improved |
| use Laplacian filter instead of random projection for H (ablation) | HarmoQ SRC · Set5 x2 (SwinIR-light) | PSNR: 37.34 -> 37.98 | Improved |
| use Laplacian filter instead of Sobel/DCT/learned basis (ablation) | HarmoQ SRC · Set5 x2 (SwinIR-light) | PSNR: 37.72-37.91 -> 37.98 | Improved |
| HarmoQ vs best baseline 2DQuant at 2-bit | HarmoQ · Set5 x2 (SwinIR-light, 2-bit) | PSNR: 36.00 -> 36.46 | Improved |
| HarmoQ vs DOBI at 3-bit on HAT-S Urban100 x4 | HarmoQ · Urban100 x4 (HAT-S, 3-bit) | PSNR: 26.89 -> 26.71 | Hurt |
| HarmoQ vs DOBI at 3-bit on HAT-S Set5 x2 | HarmoQ · Set5 x2 (HAT-S, 3-bit) | PSNR: 37.77 -> 38.35 | Improved |
Closest papers by meaning
- Outlier-Aware Post-Training Quantization for Image Super-Resolution (ICCV 2025 · similarity 0.90)
- QuantSR+: Pushing the Limit of Quantized Image Super-Resolution Networks (TPAMI 2026 · similarity 0.89)
- 2DQuant: Low-bit Post-Training Quantization for Image Super-Resolution (NeurIPS 2024 · similarity 0.88)
- PassionSR: Post-Training Quantization with Adaptive Scale in One-Step Diffusion based Image Super-Resolution (CVPR 2025 · similarity 0.88)
- AdaBM: On-the-Fly Adaptive Bit Mapping for Image Super-Resolution (CVPR 2024 · similarity 0.87)
- RAP-SR: RestorAtion Prior Enhancement in Diffusion Models for Realistic Image Super-Resolution (AAAI 2025 · similarity 0.85)
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.
