inference-visualization
Run Detectron2 inference, inspect model outputs, and visualize predictions safely without depending on the source demo script.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- inference-visualization
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Run Detectron2 inference, inspect model outputs, and visualize predictions safely without depending on the source demo script.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/inference-workflows.md6.1 KB
- 参考文档references/structures-and-outputs.md6.6 KB
- 参考文档references/troubleshooting.md5.7 KB
- 脚本scripts/demo_command_builder.py5.8 KB
- 脚本scripts/visualize_json_schema_check.py10.1 KB
03 SKILL.md 正文
Detectron2 Inference & Visualization
Use this sub-skill when a task asks to run Detectron2 inference, call DefaultPredictor, call a built model directly, inspect Instances/Boxes/masks, visualize predictions, plan demo-style image/video/webcam commands, or validate COCO-style prediction JSON before drawing it.
Fast Routing
- For simple single-image inference, read references/inference-workflows.md and use
DefaultPredictor(cfg)only when weights are local or downloads are intentional. - For batch/direct model calls, read references/inference-workflows.md and use
build_model(cfg),DetectionCheckpointer(model).load(...),model.eval(), andmodel([input_dict])[0]. - For output parsing, field checks, JSON conversion, and visualization requirements, read references/structures-and-outputs.md.
- For broken demo imports, OpenCV issues, CPU/CUDA mismatches, empty predictions, metadata labels, or output side effects, read references/troubleshooting.md.
- Build a safe demo-style command without executing it through scripts/demo_command_builder.py.
- Validate COCO-style prediction JSON without drawing or importing Detectron2 through scripts/visualize_json_schema_check.py.
Safe Defaults
- Set inference device before model or predictor construction: Yacs configs use
cfg.MODEL.DEVICE = "cpu"or the CLI overrideMODEL.DEVICE cpu. - Do not instantiate
DefaultPredictorduring no-download validation ifcfg.MODEL.WEIGHTSpoints to a URL ordetectron2://URI; it loads weights during construction. - Treat OpenCV images as BGR, but pass RGB images to
Visualizer; convert withimage_rgb = image_bgr[:, :, ::-1]. - Move predictions to CPU before visualization:
instances = outputs["instances"].to("cpu"). - Check
instances.has("pred_boxes"),instances.has("scores"), and metadata class names before assuming labels or boxes exist.
Boundaries
This sub-skill owns inference calls, model input/output formats, Instances, Boxes, BoxMode, BitMasks, PolygonMasks, ImageList, Visualizer, ColorMode, demo-style flags, confidence thresholds, CPU overrides, and prediction JSON sanity checks.
Route model-zoo config/checkpoint selection to ../configuration-model-zoo/. Route dataset registration, catalog metadata, and dataset dict validation to ../data-datasets/. Route training, evaluation loops, evaluators, and metrics to ../training-evaluation/. Route TorchScript, tracing, Caffe2, and deployment packaging to ../deployment-export/.
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).