You can deploy this model and start earning money today!

ortizeg/
basketball-rt-detrv2-m-640

Name: basketball-rt-detrv2-m-640
Availability: InStock

Views

No views yet

rt-detrv2-m-640

RT-DETRv2-M (ResNet-34-vd backbone, r34vd) fine-tuned on the basketball dataset at 640x640. Reuses the DEIM/D-FINE deploy ONNX format and identical preprocessing (square resize, RGB, /255, antialias). The weakest model in the 7-model roster on the harness test slice.

Metrics

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

Metric	5-class	10-class
mAP@50:95	0.581	0.501
mAP@50	0.862	0.708

Preprocessing

Resize: square (alignment: none)
Normalize: div255
Channel order: RGB
Input shape: (3, 640, 640) (float32)

Provenance

Source repo: https://github.com/lyuwenyu/RT-DETR
Training config: docs/provenance/configs/rtdetrv2_m/rtdetrv2_basketball_m.yml
Hardware: vast.ai RTX 3090 (single GPU)
Command: python tools/train.py -c docs/provenance/configs/rtdetrv2_m/rtdetrv2_basketball_m.yml -t rtdetrv2_r34vd_120e_coco_ema.pth --use-amp --seed 0

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("rt-detrv2-m-640")
weights_path = download_weights(card)

ortizeg/basketball-rt-detrv2-m-640