feat(cog-person-count): train count_v1.safetensors — honest v0.0.1 (ADR-103)#695
Merged
Conversation
…DR-103) Phase 2 of ADR-103: trained count head on the existing 1,077 paired samples (the same data that produced pose_v1 yesterday). Honest result: 65.1% eval accuracy / 100% within ±1 / MAE 0.349 on the held-out time-window. Per-class: 100% on "empty room" / 0% on "1 person". The model overfit by epoch 100 (train_acc → 1.0, eval_loss climbed 0.67 → 7.8) and the "best" checkpoint is the snapshot that happened to predict the eval window's class distribution (140/215 = 65.1%, matches eval_acc exactly). Confidence head Spearman = 0.023 ⇒ uncalibrated. Same data-bound failure mode as pose_v1 (#645), bounded by single-session training data; same fix path (multi-room). What v0.0.1 still validates end-to-end: * PyTorch → safetensors → Candle Rust loads cleanly on first try. `cog-person-count health` reports `backend: candle-cpu` and emits real per-frame predictions instead of the stub backend's hard-coded {1 person, 0 confidence}. Architecture parity between train-count.py and src/inference.rs::CountNet is bit-exact. * ONNX export bit-clean (16 KB, opset 18, dynamic batch axis). * Training wall time: 5.6 s for 400 epochs on RTX 5080. * Binary size unchanged (2.36 MB stripped), model loads via mmap at runtime. This commit ships: * scripts/align-ground-truth.js: extended to emit n_persons_mode + n_persons_max per window so the training pipeline has count labels. Backwards-compatible (additive fields). * scripts/train-count.py: new — mirrors CountNet architecture exactly, loads paired.jsonl, trains 400 epochs with CE+BCE+Brier loss, exports safetensors + ONNX + per-epoch JSON. * v2/.../cog/artifacts/{count_v1.safetensors,count_v1.onnx, count_train_results.json}: the trained artifacts. * v2/.../cog/README.md: Status table updated with the v0.0.1 numbers + an Honest Caveat section explaining the data-bound result. * docs/benchmarks/person-count-cog.md: new — full v0.0.1 benchmark log mirroring the format docs/benchmarks/pose-estimation-cog.md established. Includes comparison to ADR-103 v0.1.0 acceptance gates and per-class breakdown. Still pending: * `run` subcommand wiring (long-running polling loop, same as pose) * Cross-compile + sign + GCS upload (mirror of pose cog pipeline) * Live install on cognitum-v0 * v0.2.0: re-train on multi-room data, LoRA per-room adapters, Stoer-Wagner min-cut clip in fusion stage
ruvnet
added a commit
that referenced
this pull request
May 21, 2026
…al (#697) Phase 4 of ADR-103. Adds the long-running polling loop so the cog's fourth verb (`run`) does real work, completing the ADR-100 runtime contract end-to-end: cog-person-count version → "person-count 0.3.0" cog-person-count manifest → JSON skeleton cog-person-count health → loads weights + 1-shot infer + emit cog-person-count run --config → long-running per-frame emit ← THIS What ships: * src/runtime.rs (new) — `run_loop` polls sensing_url every poll_ms, slides a [56, 20] CSI window, runs InferenceEngine::infer, emits publisher::person_count events. Same shape as cog-pose-estimation::runtime — fetch_frame extracts amplitudes from `snapshot.nodes[0].amplitude[]`, fails open on connect errors with a WARN log rather than crashing. * src/lib.rs — registers the runtime module. * src/main.rs — cmd_run now loads RunConfig from a JSON file, builds the InferenceEngine (with weights if cfg.model_path is set, otherwise auto-discover), emits a run.started event, and hands off to the Tokio multi-thread runtime's block_on(run_loop). Single-node fusion is a no-op for N=1 today; v0.2.0 will append predictions from sibling nodes and call fusion::fuse_confidence_weighted before emit. Verified locally: cargo check -p cog-person-count --no-default-features → clean cargo test -p cog-person-count → 15/15 pass (no regressions) cargo build -p cog-person-count --release → 2.36 MB unchanged ./cog-person-count run --config bad-config.json: line 1: {"event":"run.started","fields":{"cog":"person-count", "sensing_url":"http://127.0.0.1:9999/...",poll_ms:100, "model_path":"(auto-discover)"}} line 2: WARN sensing-server fetch failed error=Connection Failed: Connect error: actively refused (loop alive — exits cleanly on SIGTERM, no crash, no NaN) Also adds a "Relationship to the in-process score_to_person_count heuristic" section to cog/README.md explaining the dual-emitter design (sensing-server keeps emitting the PR #491 slot heuristic; the cog runs out-of-process and emits person.count events from the learned model). Operators choose by installing the cog or not — no sensing-server rebuild required. ADR-103 §"Migration" status: 1. Land ADR + scaffold ........... done (#693, #694) 2. Train count_v1 ................ done (#695) 3. Cross-compile + sign + GCS .... done (#696) 4. Server-side wiring ............ done — out-of-process design means no rewire needed; this cog is the wiring. 5. v0.2.0 multi-room + LoRA ...... data-bound (#645)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Phase 2 of ADR-103. Trained count head on the existing 1,077 paired samples (same data that produced
pose_v1yesterday).Honest result
Model overfit by epoch ~100; the 'best' checkpoint predicts the eval-window class distribution rather than a real classifier. Same data-bound failure mode as
pose_v1(#645). v0.0.1 ships the pipeline + a working artifact + honest numbers; usable counts wait on multi-room paired data.What v0.0.1 still validates
cog-person-count healthreportsbackend: candle-cpu(not stub), architecture parity bit-exact.Files
scripts/align-ground-truth.js— extended to emitn_persons_mode+n_persons_maxper window. Backwards-compatible additive fields.scripts/train-count.py— new. MirrorsCountNetexactly; CE+BCE+Brier loss; safetensors+ONNX export.v2/.../cog/artifacts/{count_v1.safetensors, count_v1.onnx, count_train_results.json}— the artifacts.v2/.../cog/README.md— Status updated with v0.0.1 numbers + honest-caveat section.docs/benchmarks/person-count-cog.md— new benchmark log mirroring the pose-cog format.🤖 Generated with claude-flow