Research map / Image generation / Image restoration
gQIR: Generative Quanta Image Reconstruction
Aryan Garg, Sizhuo Ma, Mohit Gupta
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 latent space alignment loss (ablation) | gQIR · Stage 1 VAE (512px) | PSNR: 10.39 -> 24.78 | Improved |
| add deterministic mean encoding (ablation) | gQIR · Stage 1 VAE (512px) | PSNR: 20.56 -> 24.78 | Improved |
| add stage 2 adversarial LoRA U-Net (ablation) | gQIR · video test set (256px) | PSNR: 20.038 -> 24.114 | Improved |
| add stage 2 adversarial LoRA U-Net (ablation) | gQIR · video test set (256px) | E_warp: 9.088 -> 8.508 | Hurt |
| add stage 3 latent burst FusionViT (ablation) | gQIR · video test set (256px) | PSNR: 24.114 -> 27.630 | Improved |
| gQIR vs finetuned Restormer/NAFNet on distortion (ablation) | gQIR · Single 3-bit frame, Monochrome | PSNR: 28.728 -> 27.281 | Hurt |
| gQIR (generative prior) vs ft-NAFNet on perceptual quality | gQIR · Single 3-bit frame, Color | MUSIQ: 36.732 -> 42.038 | Improved |
| Burst-gQIR vs QuDI on I2-2000fps | gQIR · I2-2000fps | PSNR: 28.641 -> 30.811 | Improved |
| Burst-gQIR vs QUIVER on XD | gQIR · XD burst | PSNR: 20.096 -> 30.331 | Improved |
| Burst-gQIR vs QUIVER on XVFI SSIM | gQIR · XVFI (1000 fps) | SSIM: 0.751 -> 0.712 | Hurt |
| use SD3.5 VAE (4x larger latent) instead of SD2.1 (ablation) | gQIR · color SPAD | text legibility | Improved |
Closest papers by meaning
- bit2bit: 1-bit quanta video reconstruction via self-supervised photon prediction (NeurIPS 2024 · similarity 0.90)
- Generative Quanta Color Imaging (CVPR 2024 · similarity 0.89)
- When Latents Forget Pixels: Restoring Fidelity in Diffusion Transformer Super-Resolution (NeurIPS 2026 · similarity 0.86)
- Consistency Posterior Sampling for Diverse Image Synthesis (CVPR 2025 · similarity 0.86)
- Single Shot HDR Recovery via a Video Diffusion Prior (NeurIPS 2026 · similarity 0.86)
- ViDiT-Q: Efficient and Accurate Quantization of Diffusion Transformers for Image and Video Generation (ICLR 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.
