
thanks for your good job, i meet a question about train on my datasets (like kitti type). the checkpoints i saved can not detect anything, but your pretrain-model is pretty good, I find the format of checkpoint (two types, but pth) is not same, maybe it because the version of torch is different, and the val loos like the picture, can you help me solve it? thanks !