AGENT SKILL

interpolation

Use ECCV2022-RIFE image-pair, video, and PNG-sequence interpolation CLIs safely, including checkpoints, scales, ratios, and output handling.

License 许可
MIT
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
interpolation
license
MIT
metadata
{ "disco-role": "operating" }
description
Use ECCV2022-RIFE image-pair, video, and PNG-sequence interpolation CLIs safely, including checkpoints, scales, ratios, and output handling.
disable-model-invocation
true

02 资源文件 4

03 SKILL.md 正文

ECCV2022-RIFE interpolation

Use this sub-skill when the user asks to interpolate two frames, run RIFE on a video, make 2X/4X/16X slow motion, use an arbitrary --ratio, process a numbered PNG frame directory, tune UHD/4K --scale, create montage output, preserve or explain audio transfer, or diagnose RIFE checkpoint layout for inference.

Do not use this sub-skill for official benchmark metrics, dataset evaluation, training, Vimeo checkpoint production, or architecture changes. Route benchmarks/metrics to the repository evaluation guidance and route training/checkpoint production to the repository training guidance.

Operating procedure

  1. Confirm the user has a source checkout with dependencies available. The inference scripts are source scripts, not installed console entry points.

  2. Confirm a checkpoint directory. The default is train_log; a non-default directory must be passed with --model. For the current source fallback, expect flownet.pkl inside the checkpoint directory.

  3. Pick the workflow:

    • image pair: inference_img.py with --img left right, --exp or --ratio;
    • video file: inference_video.py with --video input.mp4;
    • numbered PNG frames: inference_video.py with --img frames_dir and numeric filenames such as 0.png, 1.png, ...;
    • UHD/4K: add --UHD or an explicit --scale such as 0.5.
  4. Use the bundled command builder for safe command construction and optional input/checkpoint validation before proposing a long inference run:

    python sub-skills/interpolation/scripts/interpolation_command_builder.py --help
    
  5. Before running inference, tell the user where outputs will be written. Image interpolation writes output/img*.png or output/img*.exr relative to the current working directory. Video PNG output writes vid_out/*.png; video-file output writes the requested --output or a derived *_2X_*fps.<ext> / *_4X_*fps.<ext> style path.

  6. If the user requests CUDA, FP16, or speed claims, state that functional inference can fall back to CPU but is much slower. --fp16 is only appropriate on CUDA devices with suitable Tensor Core support.

Required references

Key boundaries and caveats

  • Do not promise verified HD-model variants beyond the current source evidence. The current inference scripts attempt HD imports first, but this checkout does not provide the active model.RIFE_HD* import paths; ordinary use falls back to model.RIFE.
  • Do not claim pretrained weights are bundled. Checkpoints are external assets and must be provided by the user.
  • Do not run long video interpolation automatically. Build and validate the command first, then ask for approval if execution is potentially expensive or mutating.
  • Do not use --skip as an active static-frame optimization; the current script prints that this flag is abandoned.

04 同仓库更多 Skills

VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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).

★ 351