Back to Leaderboard

YOLOX-M

yolox

one-stage detector with CSPDarknet backbone

Parameters

25.3M

FLOPs

73.8G

Input Size

640px

License

Apache-2.0

Architecture

Type

one-stage

Backbone

CSPDarknet

Neck

PAFPN

Head

Decoupled

Benchmark Results
Performance on COCO val2017 across different hardware configurations
HardwaremAP@50-95FPSLatencyVRAM
NVIDIA A100 (TensorRT FP16)46.9%121.98.2ms1154 MB
NVIDIA T4 (TensorRT FP16)46.8%42.723.4ms1210 MB
CPU (ONNX Runtime)47.0%5.0200.1ms1218 MB
Speed Breakdown (A100 TensorRT)
End-to-end latency breakdown showing preprocessing, inference, and postprocessing times
1.2ms
4.8ms
2.3ms
Preprocess
Inference
Postprocess (NMS)
Usage with LibreYOLO
from libreyolo import YOLO

# Load model
model = YOLO.from_pretrained("https://huggingface.co/Libre-YOLO/yolox-m")

# Run inference
results = model.predict("image.jpg")

# Process results
for box in results.boxes:
    print(f"Class: {box.cls}, Confidence: {box.conf:.2f}")
balancedmegvii