artur7236 commited on
Commit
517f080
·
verified ·
1 Parent(s): 9aee863

Candidate v4: bin conf .5, stain conf .25

Browse files
Files changed (2) hide show
  1. README.md +1 -1
  2. config.json +1 -1
README.md CHANGED
@@ -14,4 +14,4 @@ Detector for the element `manako/Detect-room-dirtiness-detection`. Classes in or
14
 
15
  - `model.onnx`: YOLO11n, 640x640 input, box-only head, CPU onnxruntime through `miner.py`.
16
  - Fine-tuned from the Manako baseline (AGPL-3.0) on public validator rounds and generated images. Labels of the generated images are the consensus of three public models: Targaryen101/dirtiness_01, ninja990621/ScoreVision, dan2143/ScoreVision21. Final weights average two seeds and blend them 0.75 with the baseline.
17
- - Post-processing in `config.json`: per-class confidence and NMS, a containment rule, one box at most for spill and stain.
 
14
 
15
  - `model.onnx`: YOLO11n, 640x640 input, box-only head, CPU onnxruntime through `miner.py`.
16
  - Fine-tuned from the Manako baseline (AGPL-3.0) on public validator rounds and generated images. Labels of the generated images are the consensus of three public models: Targaryen101/dirtiness_01, ninja990621/ScoreVision, dan2143/ScoreVision21. Final weights average two seeds and blend them 0.75 with the baseline.
17
+ - Post-processing in `config.json`: per-class confidence (bin .5, stain .25) and NMS, a containment rule, one box at most for spill and stain, size rules for small bins and floor-wide spills.
config.json CHANGED
@@ -1 +1 @@
1
- {"classes": ["liquid spillage", "bin", "litter", "mould", "stain"], "imgsz": 640, "conf": [0.15, 0.2, 0.35, 0.6, 0.15], "nms_iou": [0.3, 0.45, 0.3, 0.45, 0.45], "contain_drop": [0.7, 0.7, 0.7, 0.7, 0.7], "max_per_class": [1, 99, 99, 99, 1], "small_bin": [0.5, 0.6], "big_spill": [20, 0.5], "weak_image": 0.35, "max_det": 100, "threads": 2, "model": "model.onnx"}
 
1
+ {"classes": ["liquid spillage", "bin", "litter", "mould", "stain"], "imgsz": 640, "conf": [0.15, 0.5, 0.35, 0.6, 0.25], "nms_iou": [0.3, 0.45, 0.3, 0.45, 0.45], "contain_drop": [0.7, 0.7, 0.7, 0.7, 0.7], "max_per_class": [1, 99, 99, 99, 1], "small_bin": [0.8, 0.6], "big_spill": [20, 0.5], "weak_image": 0.35, "max_det": 100, "threads": 2, "model": "model.onnx"}