qianqianderizi
commented
4 years ago I used the CenterNet model to train my data in eight categories.After changing the parameters related to the category and making the format conversion from XML to json, the iteration can be trained normally.However, when I tried to run test with the trained model, I kept misreporting CUDA out of memory
I made the following changes: 1）models/CenterNet-52.py或者models/CenterNet-104.py(看你要训练哪一个模型，下面省略)，中第132行 out_dim =80 ==> out_dim =8 (类别数) 2) config/CenterNet-52.py第45行 “categories":80 ==> “categories":8（类别数） 3）db/detection.py 第8行self._configs[ “categories"] =80 ==> self._configs[ “categories"] = 8 (类别数) 4）db/detection.py 第72行 num_classes=81 ==> num_classes=9 （类别数+1）
and my pull,push,reg loss =0 when train.
Has anyone encountered similar problems and solved them? Help me!
我也是，怎么解决呢？annotations也没啥问题呀
I used the CenterNet model to train my data in eight categories.After changing the parameters related to the category and making the format conversion from XML to json, the iteration can be trained normally.However, when I tried to run test with the trained model, I kept misreporting CUDA out of memory
I made the following changes: 1）models/CenterNet-52.py或者models/CenterNet-104.py(看你要训练哪一个模型，下面省略)，中第132行 out_dim =80 ==> out_dim =8 (类别数) 2) config/CenterNet-52.py第45行 “categories":80 ==> “categories":8（类别数） 3）db/detection.py 第8行self._configs[ “categories"] =80 ==> self._configs[ “categories"] = 8 (类别数) 4）db/detection.py 第72行 num_classes=81 ==> num_classes=9 （类别数+1）
and my pull,push,reg loss =0 when train.
Has anyone encountered similar problems and solved them? Help me!