We are using Python 3+ for this project. Also, it is recommended to run this code with available GPUs. For development, we used a nvidia GeForce GTX 1050 Ti.
The following libraries are used in this project.
| Library | Version |
|---|---|
| h5py | 2.7.1 |
| imutils | 0.4.3 |
| Keras | 2.1.1 |
| matplotlib | 2.1.1 |
| numpy | 1.14.0 |
| opencv-contrib-python | 3.3.0.10 |
| opencv-python | 3.3.0.10 |
| scikit-image | 0.13.0 |
| scikit-learn | 0.19.1 |
| scipy | 1.0.0 |
| tensorflow-gpu | 1.4.0 |
| tensorflow-tensorboard | 0.4.0rc3 |
The first step is to build the training/validation/test partitions. In order
to do this, place the images in the folder data/cells and execute the command
python3 build_partitions.py
Then, use the following command to train the model. All the parameters are
optional. If you want to re-train a model with warm-start initialization, use
the --output argument to specify the path to the weights to update.
This command will create the model file.
python3 train.py --img-path data/partitions/ --epochs 100 --output output/models/inception.h5
Then, in order to get the predictions. Use the script predict.py. The path to
the images is specified in the argument --images where you can specify either
a single image or a folder with a set of images. The output will be a line per
image with the filename and the probabilistic prediction. The model path is
specified using the --weights parameter.
python3 train.py --images data/imgs/ --weights output/models/inception.h5