Open netanel-s opened 5 years ago
The object detection part only supports the use of the B1 model as the backbone.Because the training of the A1 model is not completed. Also very embarrassed that there are some bugs in the current code. I implemented a new version, but I have encountered some problems in mixing precision training yet.
Hi @rootkitchao . Is there an expected update any time soon? Thanks a lot!
Hi. First of all, thanks for sharing your implementation and trained models! Please be aware that there's an inconsistency in the A1 model between the option of FPN min level in the explanation to the FPN levels taken from the backbone (layer_5, layer_8, layer_13, layer_19 vs 'layer_4', 'layer_7', 'layer_13', 'layer_18'). In the B1 model both are layer_5, layer_8, layer_13, layer_19.
Thanks.