The code for ACM MMSports'22 Instance Segmentation challenge. The details are presented in our technical report.

The code is based on mmdetection and CBNetV2.

ACM MMSports'22 Instance Segmentation second challenge

setup environments

sh pre.sh

train model with multi-gpus

nohup bash tools/dist_train.sh configs/mmsports2022/exp02.py 4 >> /lengyu.yb/logs/mmsports2022/exp07.log 2>&1 &

swa model with multi-gpus

nohup bash tools/dist_train.sh configs/mmsports2022/exp02_swa.py 4 >> /lengyu.yb/logs/mmsports2022/exp07_swa.log 2>&1 &

Our training logs and config are shown in weights folder

inference

bash tools/dist_test.sh configs/mmsports2022/exp02.py /lengyu.yb/logs/mmsports2022/exp07_swa/swa_model_148_mms.pth 4

generate standard submission format

python scripts/gen_standard_submission.py

evaluate on test set

python evaluate.py /lengyu.yb/datasets/segmentation/MMSports22/basketball-instants-dataset/annotations/test.json ./submission/submission.segm json ./