yjh0410 62a5b79e6a update		2 rokov pred
..
README.md	578dc295f4 redesign RTCDet	2 rokov pred
build.py	578dc295f4 redesign RTCDet	2 rokov pred
loss.py	578dc295f4 redesign RTCDet	2 rokov pred
matcher.py	578dc295f4 redesign RTCDet	2 rokov pred
rtcdet.py	578dc295f4 redesign RTCDet	2 rokov pred
rtcdet_backbone.py	578dc295f4 redesign RTCDet	2 rokov pred
rtcdet_basic.py	62a5b79e6a update	2 rokov pred
rtcdet_head.py	578dc295f4 redesign RTCDet	2 rokov pred
rtcdet_neck.py	578dc295f4 redesign RTCDet	2 rokov pred
rtcdet_pafpn.py	62a5b79e6a update	2 rokov pred
rtcdet_pred.py	578dc295f4 redesign RTCDet	2 rokov pred

RTCDet-v2: My Second Empirical Study of Real-Time Convolutional Object Detectors.

Model	Scale	Batch	AP^{test 0.5:0.95}	AP^test 0.5	AP^val 0.5:0.95	AP^val 0.5	FLOPs ^(G)	Params ^(M)	Weight
RTCDet-N	640	8xb16
RTCDet-T	640	8xb16
RTCDet-S	640	8xb16
RTCDet-M	640	8xb16
RTCDet-L	640	8xb16
RTCDet-X	640

For training, we train my RTCDet series series with 300 epochs on COCO.
For data augmentation, we use the large scale jitter (LSJ), Mosaic augmentation and Mixup augmentation, following the setting of YOLOX, but we remove the rotation transformation which is used in YOLOX's strong augmentation.
For optimizer, we use AdamW with weight decay 0.05 and base per image lr 0.001 / 64.
For learning rate scheduler, we use linear decay scheduler.
Due to my limited computing resources, I can not train RTCDet-X with the setting of batch size=128.

Model	Scale	Batch	AP^{test 0.5:0.95}	AP^test 0.5	AP^val 0.5:0.95	AP^val 0.5	FLOPs ^(G)	Params ^(M)	Weight
RTCDet-P	320	8xb16
RTCDet-P	416	8xb16
RTCDet-P	512	8xb16
RTCDet-P	640	8xb16