evaluation
Evaluate Facenet embeddings with LFW pair files, ROC/VAL/FAR metrics, and fixed-standardization checks.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- evaluation
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Evaluate Facenet embeddings with LFW pair files, ROC/VAL/FAR metrics, and fixed-standardization checks.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/metrics-reference.md1.5 KB
- 参考文档references/troubleshooting.md1.2 KB
- 参考文档references/workflows.md1.7 KB
- 脚本scripts/build_lfw_validation_command.py1.5 KB
- 脚本scripts/validate_lfw_pairs.py1.8 KB
03 SKILL.md 正文
Facenet Evaluation
Use this sub-skill when the task involves validate_on_lfw, LFW pairs, face verification metrics, training-time LFW validation, or interpreting Facenet accuracy/VAL/AUC/EER output.
When to read
- The user wants to run or debug LFW validation.
- A pair file has skipped pairs, missing images, or unexpected fold behavior.
- The task mentions
distance_metric,subtract_mean,use_flipped_images,use_fixed_image_standardization, ROC, validation rate, FAR, AUC, or EER. - The user asks how to compare new model accuracy with README pretrained model numbers.
Inputs and prerequisites
- An aligned LFW-style directory where each identity has files such as
Name/Name_0001.jpgor.png. - A pairs file with same/different identity rows. Validate it with
scripts/validate_lfw_pairs.py. - A model path accepted by Facenet. Read
../model-export-and-checkpoints/SKILL.mdif model loading fails. - A TensorFlow 1.x Facenet environment.
Workflow
- Validate the LFW directory and pairs file.
- Build a validation command with
scripts/build_lfw_validation_command.py. - For 2018 README models, include
--use_fixed_image_standardizationunless evidence shows the model expects per-image standardization. - Ensure
2 * number_of_pairs * number_of_flipsis divisible by--lfw_batch_size; the source script asserts this. - Run validation in a bounded environment and capture accuracy, validation rate, AUC, and EER lines.
- Read
references/metrics-reference.mdbefore comparing thresholds or accuracy across datasets.
Common command shape
python scripts/build_lfw_validation_command.py LFW_ALIGNED_DIR MODEL_PATH --lfw-pairs pairs.txt --lfw-batch-size 100 --lfw-nrof-folds 10 --use-fixed-image-standardization
The builder emits a module-style validate_on_lfw command.
Record the result
Always record the model path, checkpoint/frozen-graph form, pairs file, aligned-data root, image size, batch size, fold count, distance metric, mean subtraction, flip setting, fixed-standardization setting, and the printed accuracy/VAL/FAR/AUC/EER values. Without these fields, a later run is not comparable to the README benchmark.
Related routes
- Dataset alignment and pair-file layout:
../data-and-alignment/SKILL.md. - Training scripts that run LFW validation each epoch:
../training/SKILL.md. - Frozen graph/checkpoint path issues:
../model-export-and-checkpoints/SKILL.md.
Troubleshooting
Read references/troubleshooting.md for skipped pairs, batch-size assertions, fixed-standardization mismatch, insufficient folds, and metric interpretation pitfalls.
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).