video-restoration
KAIR VRT and RVRT video restoration testing and training guidance for video super-resolution, deblurring, denoising, frame interpolation, and space-time SR.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- video-restoration
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- KAIR VRT and RVRT video restoration testing and training guidance for video super-resolution, deblurring, denoising, frame interpolation, and space-time SR.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/troubleshooting.md8.9 KB
- 参考文档references/video-restoration-workflows.md9.7 KB
- 参考文档references/vrt-rvrt-task-reference.md8.5 KB
- 脚本scripts/build_video_restoration_command.py6.4 KB
03 SKILL.md 正文
KAIR video restoration router
Use this sub-skill when the task is about KAIR VRT or RVRT video workflows: video super-resolution, video deblurring, video denoising, video frame interpolation, space-time video SR, VRT/RVRT task IDs, video tiling/OOM, VRT or RVRT training configs, or RVRT custom CUDA extension failures.
Before giving commands, assume the user is in their own KAIR checkout and has
installed PyTorch plus KAIR's requirement.txt dependencies. Treat VRT/RVRT
full inference and training as CUDA-first workflows; RVRT's guided deformable
attention path requires a working CUDA-capable PyTorch build, nvcc, and
ninja for the custom extension.
Read these bundled references
references/vrt-rvrt-task-reference.mdfor the VRT001-009and RVRT001-006task namespaces, checkpoint names, default test folders, and training config paths.references/video-restoration-workflows.mdfor quick testing, custom-folder commands, training command patterns, data layout expectations, tile semantics, auto-download caveats, and metric/output behavior.references/troubleshooting.mdwhen CUDA extension builds, missing weights, missing data, Vimeo preparation, DDP checkpoint/static-graph resumes,num_workers, or OOM are involved.scripts/build_video_restoration_command.pyto print safe KAIR VRT/RVRT test commands without importing KAIR, downloading files, or launching inference.
Route to siblings for non-owned work
- Route video dataset regrouping, LMDB creation, meta-info validation, and
destructive preparation scripts to
../data-preparation/SKILL.md. - Route generic image restoration testing to
../image-testing/SKILL.md. - Route generic image training and non-video KAIR configs to
../image-training/SKILL.md. - Route shared checkpoint acquisition policy to the root model-zoo/download guidance when available; VRT/RVRT test scripts can auto-download missing weights and some testsets, so call that out before running them.
Do not ask future agents to inspect the original KAIR source or docs for these workflows. The operational task tables, command patterns, caveats, and failure modes are distilled into this sub-skill's bundled references.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).