Jens/geointel
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Record Tower YOLO runtime validation
GeoIntel CI / docs-smoke (push) Has been cancelled
Details
GeoIntel CI / contract-smoke (push) Has been cancelled
Details

Browse Source
This commit is contained in:
Codex
2026-07-09 02:31:54 +02:00
parent f9493479c7
commit 797161d760
3 changed files with 9 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/CODEX_EXECUTION_LOG.md
+7 -2
View File
@@ -5698,12 +5698,17 @@ Open:
- 16 samples, 13 positive samples, 9 background samples
- 11,757 labels, 0 missing label files, 0 invalid label rows
- repeated background negative share 0.0
- Tower all-in-one rebuild from commit `f949347` with `GEOINTEL_INSTALL_AI=true`:
- live migration smoke passed.
- browser runtime verification passed on `http://192.168.10.150:1202`.
- `/app/scripts/prepare_operator_real_data_samples.py`, `/app/scripts/export_operator_yolo_tile_dataset.py` and `/app/scripts/audit_operator_yolo_dataset_quality.py` are present in the rebuilt container.
- YOLO preflight with an existing raster tile manifest returned `status: ready`, `dependencies_available: true`, `model_file_exists: true`, `tile_paths_exist: true`, `will_download_models: false` and `will_run_inference: false`.
## Known limitations
- Some background candidates contain real GRB buildings. They are still useful as mixed rural/background samples, but the pure negative pressure currently comes mostly from Postel-bos, Lommel-heide and Arendonk-heide plus empty tiles inside sparse candidates.
- The running Tower container was updated via temporary `docker cp` for live data prep before the Dockerfile copy fix existed; a normal rebuild is needed for `/app/scripts/prepare_operator_real_data_samples.py` to exist inside the image automatically.
- The rebuilt Tower image now contains the operator scripts automatically. The generated `yolo-building-tile-uniquehardneg160` dataset is ready for a controlled training candidate, but no model has been promoted from it yet.
## Next recommended pass
- Rebuild the Tower all-in-one image, then use `yolo-building-tile-uniquehardneg160` as the next safer hard-negative training dataset candidate. Benchmark after training before changing defaults.
- Use `yolo-building-tile-uniquehardneg160` as the next safer hard-negative training dataset candidate. Benchmark after training before changing defaults.