Convert WIDER Face Dataset to Tensorflow's TFRecord format.
Convert the WIDER Face to TFRecord format.
Important: this project runs under Python3.
The TFRecord format will allow you to use the Tensorflow Object Detection API in order to train a new model based on Transfer Learning from many available pretrained models.
pip install -r requirements.txt
config.py
file with your appropiate paths.
None
. For example, if test is not going to be converted you should leave TEST_WIDER_PATH
in None
.python wider_to_tfrecord.py
tfrecord
files should be in the output folder you specified in config.py
.Testing images do not contain ground truth bounding boxes. In case you want to convert the testing set as well, the test.tfrecord
file will contain images only without bounding boxes.