Improve results

Hello !
I succeeded to train my own dataset with code GitHub - WXinlong/SOLO: SOLO and SOLOv2 for instance segmentation, ECCV 2020 & NeurIPS 2020. with soloV2 and ResNet 101
But the results are relatively bad.
How can I improve the results please ?
Thank you
Best regards

nobody has any idea ?

Please don’t bump a topic after 3 hours.
As a general advice, I would try to overfit a small data sample by playing around with some hyperparameters, and once this is done scale up the use case again.
Also, some (open) ML courses might be helpful, e.g. you might want to take a look into the FastAI course.