(Detectron's source interpretation)/detectron/lib/datasets/

1. dataset_catalog.py

When we need to train our own data (for reference: http://blog.csdn.net/meccaendless/article/details/79457330), we need to modify the contents of the file, when I use my own data for training, I'm using the VOC2007 data structure, so I modified the following part of the file:

' Voc_2007_trainval ': {

Im_dir:

_data_dir + '/voc2007/jpegimages ',

ANN_FN:

_data_dir + '/voc2007/annotations/voc_2007_trainval.json ',

Devkit_dir:

_data_dir + '/voc2007/vocdevkit2007 '

},

' Voc_2007_test ': {

Im_dir:

_data_dir + '/voc2007/jpegimages ',

ANN_FN:

_data_dir + '/voc2007/annotations/voc_2007_test.json ',

Devkit_dir:

_data_dir + '/voc2007/vocdevkit2007 '

},

Modify to the following form:

' Voc_2007_trainval ': {

Im_dir:

_data_dir + '/voc2007/jpegimages ',

ANN_FN:

_data_dir + '/voc2007/annotations/pascal_trainval2007.json ',

Devkit_dir:

_data_dir + '/voc2007/vocdevkit2007 '

},

' Voc_2007_test ': {

Im_dir:

_data_dir + '/voc2007/jpegimages ',

ANN_FN:

_data_dir + '/voc2007/annotations/pascal_test2007.json ',

Devkit_dir:

_data_dir + '/voc2007/vocdevkit2007 '

},

2. dummy_datasets.py

When we train our own data, the file is not necessarily to be modified, but when we finally test the model with infer_simple.py, we need to modify the contents of the classes inside, otherwise the label on the image is the Coco Data set label. PS: Don't forget ' __background__ '