training for Resized imgs

[chengdianJiang]

Thanks for the amazing work, I want to train my data with the size of 416*416, and tried to use ----transforms.Compose([Normalizer(), Augmenter(), Resizer()]) at line 69 in train.py, but it can not work.

[pjl1995]

Maybe your label is not appropriate. In MOT17, the sizes of imgs are not exactly the same, including 19201080 and 640480.

[chengdianJiang]

Thanks for your reply, I trained the model with MOT17, in the RandomSampleCrop, The Croped imgs have been resized to 416416,
and the total loss was around~ 1.1, but the test results is not good(I also test the MOT17 imgs with the resized 416416),. So may i ask what is the loss condition during your training work. Thanks again!