using stage 4 to training for objects holding by hand

[xiang-zhe]

Great project, THX something in hand will be detected as background sometimes, so i want to finetune the model，but now， only some images with alpha was find，so if i any trian it for stage 4，will it works？