Exact Submission Reproduction
This folder contains the exact inference artifact needed to reproduce the submitted model behavior:
run.pybest_yolo.ptrequirements.txt
What this reproduces:
- The exact inference package structure needed to run the submission logic locally.
- The exact model weight file that should be used with this
run.py.
What this does not reproduce by itself:
- The private competition leaderboard score, unless the evaluator has access to the challenge platform and hidden evaluation images.
- Full end-to-end training, because the competition dataset and product reference images are not included here.
Recommended hosting:
- Upload this folder to a Hugging Face model repository.
- Upload the source-only folder separately to GitHub if you also want a normal code repository.
Suggested repo layout on Hugging Face:
run.pybest_yolo.ptrequirements.txtREADME.mdSHA256SUMS.txt
Local run command:
python run.py --input /path/to/images --output /path/to/predictions.jsonExpected input:
- Image files named like
img_00042.jpg
Expected output:
- A JSON array of predictions in the competition format.
Notes:
- The competition sandbox uses Python 3.11, PyTorch 2.6.0, torchvision 0.21.0, ultralytics 8.1.0, numpy 1.26.4, opencv-python-headless 4.9.0.80, and ensemble-boxes 1.0.9.
- If the evaluator wants to reproduce the exact public submission package, they can zip the contents of this folder with
run.py at the zip root.
