YOLOX:

For training, we train YOLOX series with 300 epochs on COCO.
For data augmentation, we use the large scale jitter (LSJ), Mosaic augmentation and Mixup augmentation.
For optimizer, we use SGD with weight decay 0.0005 and base per image lr 0.01 / 64,.
For learning rate scheduler, we use Cosine decay scheduler.
The reason for the low performance of my reproduced YOLOX-L has not been found out yet.

Model	Backbone	Batch	Scale	AP^val 0.5:0.95	AP^val 0.5	FLOPs ^(G)	Params ^(M)	Weight
YOLOX-S	CSPDarkNet-S	8xb8	640	40.1	60.3	26.8	8.9	ckpt
YOLOX-M	CSPDarkNet-M	8xb8	640	46.2	66.0	74.3	25.4	ckpt
YOLOX-L	CSPDarkNet-L	8xb8	640	48.7	68.0	155.4	54.2	ckpt
YOLOX-X	CSPDarkNet-X	8xb8	640