onnx-and-browser
Export Segment Anything's prompt encoder and mask decoder to ONNX, quantize it, prepare matching image embeddings, and wire the model into ONNXRuntime or browser workflows.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- onnx-and-browser
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Export Segment Anything's prompt encoder and mask decoder to ONNX, quantize it, prepare matching image embeddings, and wire the model into ONNXRuntime or browser workflows.
- disable-model-invocation
- true
02 资源文件 6
- 参考文档references/api-reference.md2.6 KB
- 参考文档references/browser-demo.md3.0 KB
- 参考文档references/onnx-export.md3.4 KB
- 参考文档references/troubleshooting.md2.9 KB
- 脚本scripts/export_onnx_model.py7.2 KB
- 脚本scripts/prepare_embedding_template.py2.8 KB
03 SKILL.md 正文
ONNX and Browser Workflows
Use this sub-skill when the user asks to export SAM to ONNX, quantize a SAM ONNX model, run the mask decoder in ONNXRuntime Web, prepare a .npy image embedding for a browser demo, or diagnose ONNX/browser input shape issues.
Route First
- For prompted Python inference with
SamPredictor.predict, use../prompted-segmentation/unless ONNX export or embeddings are part of the request. - For full-image automatic mask generation with
SamAutomaticMaskGenerator, use../automatic-mask-generation/. - For browser workflows, export both a mask decoder ONNX file and image embeddings produced by the same checkpoint/model type/image preprocessing path.
Fast Paths
- Export ONNX with
python sub-skills/onnx-and-browser/scripts/export_onnx_model.py --checkpoint sam_vit_h_4b8939.pth --model-type vit_h --output sam_onnx.onnx. - Quantize for browser size/runtime with
--quantize-out sam_onnx_quantized.onnx; this requiresonnxruntimebecause quantization uses ONNXRuntime tooling. - Prepare a matching embedding by adapting
scripts/prepare_embedding_template.py; save the.npybeside the browser image asset. - In browser feeds, use input names
image_embeddings,point_coords,point_labels,mask_input,has_mask_input, andorig_im_size.
References
references/onnx-export.mdexplains export flags, tensor contracts, dynamic axes, quantization, and embedding compatibility.references/browser-demo.mdexplains browser asset layout, ONNXRuntime Web feeds, scaling, and SharedArrayBuffer headers.references/api-reference.mdsummarizesSamOnnxModel, export-script behavior, input/output names, and expected shapes.references/troubleshooting.mdmaps common export/runtime/browser errors to concrete fixes.
Bundled Scripts
scripts/export_onnx_model.pyis an adapted export helper with safe--helpbehavior and clearer optional-dependency errors.scripts/prepare_embedding_template.pyis a template for exporting a.npyimage embedding fromSamPredictorfor ONNX/browser use.
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).