The code is an unofficial pytorch implementation of SOLOv2: Dynamic, Faster and Stronger

Please check SOLOv1 for installation instructions.

Follows the same way as SOLOv1.

single GPU:

python tools/train.py configs/solov2/solov2_r101_3x.py

multi GPU (for example 8):

./tools/dist_train.sh configs/solov2/solov2_r101_3x.py 8

Trained model can be download in Google drive BaiduYun 提取码: qw4e

After training 36 epochs(3x) on the coco dataset using the resnet-101 backbone, the mAP is 39.5 on COCO test-dev2017 dataset. In the original paper, the model achieves 39.7 after 72 epochs(6x).