Bugs appearing:
1: Error after running the training script: ImportError: No module named _mask
Reference: https://blog.csdn.net/sinat_41852207/article/details/87935955
2:ImportError: numpy.core.multiarray failed to import
Reason numpy version is too low, you can execute the statement pip install -U numpy, it is best to look at the next bug
3:ImportError: cannot import name _validate_lengths
The reason is that the numpy version is too high, just install version 1.15.0. Pip install numpy == 1.15.0
4:AttributeError: ‘RoIDataLayer’ object has no attribute ‘param_str_’
Solution: In the latest caffe version, param_str_ is renamed, called param_str, and param_str_ can be changed to param_str. (Change py-faster-rcnn \ lib \ roi_data_layer \ layer.py here)
5:TypeError: slice indices must be integers or None or have an __index__ method
Modify D: \ software_engineer \ py-faster-rcnn \ lib \ rpn \ proposal_target_layer.py, add an int to force type conversion
6 Note: Delete the files in data \ cache and data \ VOCdevkit2007 \ annotations_cache before each training.