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

Add fail-closed source class and SAM fallback filters
GeoIntel release gates / Compile, test, contracts and builds (push) Canceled after 0s
Details
GeoIntel release gates / Python and npm vulnerability policy (push) Canceled after 0s
Details
GeoIntel release gates / GIS image, SBOM and container scan (push) Canceled after 0s
Details

Browse Source
This commit is contained in:
Jens
2026-07-30 04:24:35 +02:00
parent 29378bba42
commit 08422ca50c
6 changed files with 74 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/CODEX_EXECUTION_LOG.md
+5
View File
@@ -3754,12 +3754,17 @@ Validation:
- Ran the checksummed `/app/sam2_t.pt` model on the server RTX 4080 SUPER and rendered deterministic before/after label contact sheets.
- Rejected both the legacy refinement settings and a stricter trial because dense or temporally absent footprint prompts still produced visually invalid roof boxes; neither result was promoted into a training corpus.
- Hardened `scripts/refine_yolo_labels_with_sam.py` with configurable centre-shift and per-dimension ratio gates. The selected values and model checksum are persisted in `sam-refinement.json`.
- Added an optional provider-native source-class allowlist to Belgian label normalization. A V65 diagnostic retained only GRB `TYPE=1` main buildings and auditable rejection counts, reducing the three candidate AOIs from 427 to 283 source features without claiming semantic parity with PICC or UrbIS.
- Added explicit SAM fallback policies. `retain` preserves historical behavior; `drop` creates a fail-closed visible-roof-only shard and records every removed fallback.
- Tested global image-edge translation at narrow and broad radii. The broad optima jumped to unrelated repeated urban edges, so this experiment was rejected and was not added to the production corpus.
### What was tested
- `python -m pytest backend/tests/test_sam_roof_label_refinement.py -q` passed (`2 passed`).
- `python -m py_compile scripts/refine_yolo_labels_with_sam.py` passed.
- `python -m pytest backend/tests/test_sam_roof_label_refinement.py backend/tests/test_building_label_normalization.py -q` passed (`10 passed`).
- CUDA refinement completed on 706 candidate labels; the strictest trial refined 659 and retained 47 explicit fallbacks. Visual QA still failed, so the run remains evidence only.
- The V65 main-building-only/drop-fallback CUDA trial retained 134 of 466 tiled labels and dropped 332 fallbacks. Its contact sheet still contained wrong tree, ground and compound masks, so it also remains rejected evidence only.
### What remains open