YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
SN-44 TurboVision — manak0/Detect-beverage-detect miner
Detects cup, can, bottle in ScoreVision challenge frames and returns boxes in the
element's manifest class order.
Provenance and modification notice (AGPL-3.0 §5(a))
This repository conveys a work derived from Ultralytics YOLO11s, which its author licenses
under AGPL-3.0 (https://ultralytics.com/license). The licence to use, modify and propagate
it reaches us directly from the original licensors under AGPL-3.0 §10.
| item | value |
|---|---|
| upstream architecture | Ultralytics YOLO11s (AGPL-3.0), exported ONNX, opset 18, fp16 |
| weights obtained from | Realfencer/ScoreVision, revision d2b7d836e695de6c334f233ca25a0c1c6a0fc49b |
weights.onnx sha256 |
c727aad6d3d06814c4151e24ad76af31207764bc924f307cce4e4b113a22e77e |
| weights modified? | No. Byte-identical to the artifact obtained above. |
| modified on | 2026-09-07 |
What we changed (everything in this repo other than weights.onnx):
miner.py— our own implementation: raw-head decode of the[1, 7, 33600]output, our own class-aware NMS, letterboxing, greyscale/BGRA/non-uint8 input handling, auto-gamma, geometry filters, and the class remap described below.config.json— our thresholds, geometry and class mapping.chute_config.yml— our serving configuration.
Why a remap is needed. The upstream graph's own class order is
{0: cup, 1: bottle, 2: can}. The element manifest that the validator indexes cls_id through is
['cup', 'can', 'bottle'] — the order changed between residencies of this element (the prior
manifest was ['cup', 'bottle', 'can']). config.json therefore declares the model's order and
maps by name, so emitted ids are cup -> 0, can -> 1, bottle -> 2. Shipping these weights
without that remap silently scores AP 0 on two of the three classes.
Corresponding source (AGPL-3.0 §13)
This model is served over a network. The complete corresponding source for the work conveyed here
is this repository, which is public: weights.onnx, miner.py, config.json, chute_config.yml.
Upstream training code for YOLO11s is at https://github.com/ultralytics/ultralytics (AGPL-3.0).
Licence
AGPL-3.0. See LICENSE. We do not relicense, and we have not removed the license metadata key
from weights.onnx.
- Downloads last month
- 18