Research map / Video / Video segmentation and tracking
EVDI++: Event-Based Video Deblurring and Interpolation via Self-Supervised Learning
Chi Zhang, Xiang Zhang, Chenxu Jiang, Gui-Song Xia, Yu Lei
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 blurry-event loss L_B-E to the sharp-blurry loss alone (ablation) | EVDI++ self-supervised losses · ColorDVS-Real motion deblurring | PSNR: 28.84 -> 30.04 | Improved |
| add blurry-event loss L_B-E to L_S-B plus L_B-B (ablation) | EVDI++ self-supervised losses · ColorDVS-Real motion deblurring | PSNR: 31.57 -> 32.05 | Improved |
| add blurry-event loss L_B-E to blurry-blurry loss alone (ablation) | EVDI++ self-supervised losses · ColorDVS-Real motion deblurring | PSNR: 31.39 -> 12.72 | Hurt |
| add sharp-blurry loss L_S-B for brightness consistency (ablation) | EVDI++ self-supervised losses · ColorDVS-Real motion deblurring | PSNR: 12.72 -> 32.05 | Improved |
| add sharp-blurry loss L_S-B to blurry-event loss alone (ablation) | EVDI++ self-supervised losses · ColorDVS-Real motion deblurring | PSNR: 10.47 -> 30.04 | Improved |
| add blurry-blurry loss L_B-B to sharp-blurry loss alone (ablation) | EVDI++ self-supervised losses · ColorDVS-Real motion deblurring | PSNR: 28.84 -> 31.57 | Improved |
| add sharp-blurry loss L_S-B to blurry-blurry loss alone (ablation) | EVDI++ self-supervised losses · ColorDVS-Real motion deblurring | PSNR: 31.39 -> 31.57 | Improved |
| add all three losses versus L_S-B alone on frame interpolation (ablation) | EVDI++ self-supervised losses · ColorDVS-Real frame interpolation | PSNR: 31.78 -> 32.83 | Improved |
| upgrade EVDI to EVDI++ (LDR, APF, new losses) | EVDI · ColorDVS-Real motion deblurring (single frame) | PSNR: 26.94 -> 32.37 | Improved |
| replace learned fusion network with parameter-free APF and drop L_S-E | EVDI · REDS training time | seconds per epoch: 5760 -> 360 | Improved |
| scale up EVDI++ to the Large model | EVDI++ · BS-ERGB 3-SKIP | PSNR: 25.43 -> 26.53 | Improved |
| retrain baselines on the target training sets | RED-Net / EFNet · ColorDVS-Real motion deblurring (single frame) | PSNR: 20.92 -> 34.86 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- RED-Net* — RED-Net: A Recurrent Encoder–Decoder Network for Video-Based Face Alignment (IJCV 2018) · better on 4 of 4 shared results
- TimeLens — Time Lens: Event-Based Video Frame Interpolation (CVPR 2021) · better on 1 of 2 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
- Unifying Motion Deblurring and Frame Interpolation With Events (CVPR 2022 · similarity 0.94)
- A Unified Framework for Event-Based Frame Interpolation With Ad-Hoc Deblurring in the Wild (TPAMI 2024 · similarity 0.92)
- Event-Based Frame Interpolation With Ad-Hoc Deblurring (CVPR 2023 · similarity 0.91)
- Learning Event-Driven Video Deblurring and Interpolation (ECCV 2020 · similarity 0.90)
- Frequency-aware Event-based Video Deblurring for Real-World Motion Blur (CVPR 2024 · similarity 0.90)
- Event-Enhanced Blurry Video Super-Resolution (AAAI 2025 · 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.
