Research map / Image generation / Super-resolution
Q-DiT4SR: Exploration of Detail-Preserving Diffusion Transformer Quantization for Real-World Image Super-Resolution
Xun Zhang, Kaicheng Yang, Hongliang Lu, Haotong Qin, Yong Guo, 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 H-SVD (ablation) | quantized DiT4SR baseline · RealSR W4A6 | MUSIQ: 65.84 -> 67.46 | Improved |
| add VaSMP on top of H-SVD (ablation) | Q-DiT4SR · RealSR W4A6 | MUSIQ: 67.46 -> 67.72 | Improved |
| add naive global-MSE mixed precision (MP) to H-SVD (ablation) | Q-DiT4SR · RealSR W4A6 | MUSIQ: 67.46 -> 66.84 | Hurt |
| replace naive MP with VaSMP (ablation) | Q-DiT4SR · RealSR W4A6 | CLIP-IQA: 0.5335 -> 0.5671 | Improved |
| add VaTMP on top of H-SVD and VaSMP (ablation) | Q-DiT4SR · RealSR W4A4 | MANIQA: 0.4227 -> 0.4367 | Improved |
| add VaTMP on top of H-SVD and VaSMP (ablation) | Q-DiT4SR · RealSR W4A4 | LIQE: 3.091 -> 3.179 | Improved |
| add H-SVD and VaSMP to W4A4 baseline (ablation) | quantized DiT4SR · RealSR W4A4 | LIQE: 3.191 -> 3.091 | Hurt |
| increase SVD-L rank from 4 to 8 (ablation) | H-SVD · RealSR W4A6 | MUSIQ: 66.71 -> 67.72 | Improved |
| increase SVD-L rank from 8 to 32 (ablation) | H-SVD · RealSR W4A6 | MUSIQ: 67.72 -> 68.01 | No clear effect |
| increase SVD-L rank from 8 to 32 (ablation) | H-SVD · RealSR W4A6 | FLOPs (G): 208.22 -> 213.28 | Hurt |
| calibrate VaTMP on DIV2K instead of RealSR (ablation) | VaTMP · RealSR W4A4 | MANIQA: 0.4367 -> 0.4228 | No clear effect |
| add full Q-DiT4SR (H-SVD, VaSMP, VaTMP) over baseline (ablation) | DiT4SR W4A4 · RealSR W4A4 ROI | Edge MAE: 0.2757 -> 0.2235 | Improved |
Closest papers by meaning
- Q-DiT: Accurate Post-Training Quantization for Diffusion Transformers (CVPR 2025 · similarity 0.92)
- Effective Diffusion Transformer Architecture for Image Super-Resolution (AAAI 2025 · similarity 0.91)
- PTQ4DiT: Post-training Quantization for Diffusion Transformers (NeurIPS 2024 · similarity 0.91)
- PassionSR: Post-Training Quantization with Adaptive Scale in One-Step Diffusion based Image Super-Resolution (CVPR 2025 · similarity 0.90)
- TSD-SR: One-Step Diffusion with Target Score Distillation for Real-World Image Super-Resolution (CVPR 2025 · similarity 0.90)
- SinSR: Diffusion-Based Image Super-Resolution in a Single Step (CVPR 2024 · similarity 0.90)
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.
