You can deploy this model and start earning money today!

ortizeg/
basketball-yolox-m-800

Name: basketball-yolox-m-800
Availability: InStock

Views

No views yet

yolox-m-800

YOLOX-M anchor-free detector fine-tuned on the basketball dataset at an 800x800 input resolution. Reused from an earlier training run (not part of the @640 final-comparison retrain batch); its harness test-set score came from correcting the preprocessing (top-left pad-114 letterbox, BGR, no normalization) to match training, which alone moved its mAP@50:95 from 30.8 to 72.3.

Metrics

Measured on basketball-player-detection-3 (test split), via the object-detection-eval harness.

Metric	5-class	10-class
mAP@50:95	0.672	—
mAP@50	0.934	—

Preprocessing

Resize: letterbox (alignment: top_left)
Normalize: none
Channel order: BGR
Input shape: (3, 800, 800) (float32)

Provenance

Source repo: https://github.com/ortizeg/YOLOX
Command: see docs/provenance/training-runs.md#yolox-m-800

Usage

This ONNX file is one of the 7-model roster benchmarked in object-detection-eval. Load it through the registry for verified, hash-checked download and the exact preprocessing recipe above:

python

from object_detection_eval.registry import ModelRegistry, download_weights

registry = ModelRegistry.from_directory("registry")
card = registry.get("yolox-m-800")
weights_path = download_weights(card)

ortizeg/basketball-yolox-m-800