EcoNode model attribution

Model: PFCS33/waste-sort-yolo, weights/best.pt
Source: https://github.com/PFCS33/waste-sort-yolo
Revision: 0194b48b4dd8d8356be29da9523ec5282109d391
Source artifact: https://github.com/PFCS33/waste-sort-yolo/blob/0194b48b4dd8d8356be29da9523ec5282109d391/weights/best.pt

The downloaded artifact is a six-class YOLOv8n detector:
BIODEGRADABLE, CARDBOARD, GLASS, METAL, PAPER, PLASTIC.
The repository describes several models; its 14-class model metrics must not
be attributed to this six-class weight file.

Changes for EcoNode (2026-10-30): exported the original weights to static
FP32 ONNX, opset 17, 640 x 640, batch 1, without embedded NMS. No retraining.
Browser postprocessing uses confidence floor 0.10 and class-aware NMS at 0.70.
Chinese labels are translations of the original model class names.

The upstream README declares the project licensed under the MIT License.
Source license statement: https://github.com/PFCS33/waste-sort-yolo#license
Upstream credits Ultralytics, TACO and TrashNet.
Ultralytics YOLO terms: https://www.ultralytics.com/license
ONNX Runtime is MIT licensed: https://github.com/microsoft/onnxruntime/blob/main/LICENSE

These labels are visual predictions, not measurements of physical material.
The model can miss objects and assign incorrect labels with high confidence.
No lid-opening or disposal decision is made by this demo.