Research map / Video / Video segmentation and tracking
NanoVSR: Towards Real-Time Video Super-Resolution on Edge Devices
Filip Pawlicki, Marcel Kańduła, Marcin Pucek, Kamil Dobies
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 |
|---|---|---|---|
| fuse multi-branch blocks into single 3x3 conv at inference (ablation) | NanoVSR-226k · H100 runtime | runtime (ms/frame): 3.471 -> 1.910 | Improved |
| fuse multi-branch blocks into single 3x3 conv at inference (ablation) | NanoVSR-226k · REDS4 | PSNR (dB): 28.23 -> 28.23 | No clear effect |
| train with multi-branch reparameterizable blocks instead of single branch (ablation) | NanoVSR-226k · REDS4 | PSNR (dB): 28.13 -> 28.23 | Improved |
| train with multi-branch reparameterizable blocks instead of single branch (ablation) | NanoVSR-226k · Vid4 | SSIM: 0.7296 -> 0.7252 | Hurt |
| add Vimeo-90K short-sequence pre-training phase (ablation) | NanoVSR-226k · Vid4 | PSNR (dB): 25.17 -> 25.26 | Improved |
| add Vimeo-90K short-sequence pre-training phase (ablation) | NanoVSR-226k · REDS4 | PSNR (dB): 28.23 -> 28.23 | No clear effect |
| add Vimeo-90K short-sequence pre-training phase (ablation) | NanoVSR-226k · Vimeo90K-T | PSNR (dB): 34.21 -> 34.31 | Improved |
| add SPyNet explicit alignment (ablation) | NanoVSR-226k · REDS4 | PSNR (dB): 28.23 -> 29.44 | Improved |
| add SPyNet explicit alignment (ablation) | NanoVSR-226k · H100 runtime | runtime (ms/frame): 1.910 -> 3.953 | Hurt |
| scale NanoVSR from 22k to 644k parameters | NanoVSR · REDS4 | PSNR (dB): 27.62 -> 28.64 | Improved |
| scale NanoVSR from 5.4M to 9.6M parameters | NanoVSR · REDS4 | PSNR (dB): 29.73 -> 29.90 | No clear effect |
| run on Jetson Orin NX 16GB (25W) instead of 8GB (15W) | NanoVSR-226k · Jetson 180x320 | FPS: 23.55 -> 43.86 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- BasicVSR — BasicVSR: The Search for Essential Components in Video Super-Resolution and Beyond (CVPR 2021) · better on 2 of 5 shared results
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
- BasicVSR: The Search for Essential Components in Video Super-Resolution and Beyond (CVPR 2021 · similarity 0.87)
- FlashVSR: Towards Real-time Diffusion-Based Streaming Video Super Resolution (CVPR 2026 · similarity 0.87)
- TAQ: Static-Deployable Temporal-Aware Quantization for Real-World Video Super-Resolution (ECCV 2026 · similarity 0.87)
- SAVSR: Arbitrary-Scale Video Super-Resolution via a Learned Scale-Adaptive Network (AAAI 2024 · similarity 0.87)
- Edge-SD-SR: Low Latency and Parameter Efficient On-device Super-Resolution with Stable Diffusion via Bidirectional Conditioning (CVPR 2025 · similarity 0.86)
- Arbitrary-Scale Video Super-resolution Guided by Dynamic Context (AAAI 2024 · similarity 0.86)
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.
