Research map / Image generation / Autoregressive and latent generation
NanoSD: Edge Efficient Foundation Model for Real Time Image Restoration
Subhajit Sanyal, Srinivas Soumitri Miriyala, Akshay Janardan Bankar, Manjunath Arveti, Sowmya Vajrala, Shreyas Pandith, Sravanth Kodavanti, Abhishek Ameta, Harshit Harshit, Amit Satish Unde
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 |
|---|---|---|---|
| use NanoSD-Prime (block-searched U-Net) instead of hand-tuned SD 1.5 variant (ablation) | SD 1.5 U-Net · teacher-aligned FID | taFID: 20.6 -> 10 | Improved |
| use NanoSD 2 instead of TinySD | SD 1.5 U-Net · NPU latency | latency (ms): 74 -> 27 | Improved |
| use NanoSD 2 instead of TinySD | SD 1.5 U-Net · teacher-aligned FID | taFID: 13.8 -> 10 | Improved |
| use NanoSD 1 instead of NanoSD 2 (ablation) | NanoSD · NPU latency | latency (ms): 27 -> 41 | Hurt |
| use NanoSD backbone in OSDFace | OSDFace · CelebA-Test | MACs (G): 2465 -> 479 | Improved |
| use NanoSD backbone in OSDFace | OSDFace · CelebA-Test | FID: 45.41 -> 45.92 | No clear effect |
| use NanoSD backbone in Diff-Plugin | Diff-Plugin · desnowing, dehazing, deblurring, deraining | FID: 34.30 -> 34.83 (desnowing) | No clear effect |
| use NanoSD backbone in Diff-Plugin | Diff-Plugin · restoration tasks | parameters (M): 1256 -> 712 | Improved |
| use NanoSD backbone in Marigold | Marigold · NYUv2 zero-shot depth | AbsRel: 5.5 -> 7.2 | Hurt |
| use NanoSD backbone in Marigold | Marigold · KITTI zero-shot depth | AbsRel: 9.9 -> 11.8 | Hurt |
| use NanoSD backbone in DiffBIR | DiffBIR · CelebA-Test | FID: 71.77 -> 70.89 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 4 on most shared results.
- Diff-Plugin — Diff-Plugin: Revitalizing Details for Diffusion-based Low-level Tasks (CVPR 2024) · better on 1 of 5 shared results
- OSDFace — OSDFace: One-Step Diffusion Model for Face Restoration (CVPR 2025) · better on 1 of 2 shared results
- Edge-SD-SR — Edge-SD-SR: Low Latency and Parameter Efficient On-device Super-Resolution with Stable Diffusion via Bidirectional Conditioning (CVPR 2025) · better on 1 of 1 shared result
- PocketSR — PocketSR: The Super-Resolution Expert in Your Pocket Mobiles (NeurIPS 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Tuning Real-World Image Restoration at Inference: A Test-Time Scaling Paradigm for Flow Matching Models (ECCV 2026 · similarity 0.89)
- TPGDiff : Hierarchical Triple-Prior Guided Diffusion for Image Restoration (ICML 2026 · similarity 0.89)
- GenDeg: Diffusion-based Degradation Synthesis for Generalizable All-In-One Image Restoration (CVPR 2025 · similarity 0.89)
- Edge-SD-SR: Low Latency and Parameter Efficient On-device Super-Resolution with Stable Diffusion via Bidirectional Conditioning (CVPR 2025 · similarity 0.88)
- DiffuseHigh: Training-Free Progressive High-Resolution Image Synthesis Through Structure Guidance (AAAI 2025 · similarity 0.88)
- When Latents Forget Pixels: Restoring Fidelity in Diffusion Transformer Super-Resolution (NeurIPS 2026 · 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.
