yjh0410 fd8e9e5e0c update ELANNet-v2 2 years ago
..
README.md 6f4f753562 keep training YOLOX-L 2 years ago
build.py 2eb02320fc release RTCDet-v1 & rename RTMDet to RTCDet 2 years ago
loss.py a1b69fb635 update 2 years ago
matcher.py 8d590a1158 modify RTCDet-v2 2 years ago
rtcdet_v2.py 5754e59167 debug RTCDet-v2 2 years ago
rtcdet_v2_backbone.py fd8e9e5e0c update ELANNet-v2 2 years ago
rtcdet_v2_basic.py fd8e9e5e0c update ELANNet-v2 2 years ago
rtcdet_v2_head.py fd8e9e5e0c update ELANNet-v2 2 years ago
rtcdet_v2_neck.py fd8e9e5e0c update ELANNet-v2 2 years ago
rtcdet_v2_pafpn.py fd8e9e5e0c update ELANNet-v2 2 years ago
rtcdet_v2_pred.py 2eb02320fc release RTCDet-v1 & rename RTMDet to RTCDet 2 years ago

README.md

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

Model Scale Batch APtest
0.5:0.95
APtest
0.5
APval
0.5:0.95
APval
0.5
FLOPs
(G)
Params
(M)
Weight
RTCDetv2-N 640 8xb16
RTCDetv2-T 640 8xb16
RTCDetv2-S 640 8xb16
RTCDetv2-M 640 8xb16
RTCDetv2-L 640 8xb16
RTCDetv2-X 640
  • For training, we train my RTCDetv2 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 RTCDetv2-X with the setting of batch size=128.
Model Scale Batch APtest
0.5:0.95
APtest
0.5
APval
0.5:0.95
APval
0.5
FLOPs
(G)
Params
(M)
Weight
RTCDetv2-P 320 8xb16
RTCDetv2-P 416 8xb16
RTCDetv2-P 512 8xb16
RTCDetv2-P 640 8xb16