3D-CNN_DataGenerator

DataGenerator for 3D-CNN in keras and tensorflow

This repo contains the code of data generator for 3DCNN architectures. The code was built using multiple resources but most of the content is taken from Shervine Amidi

Prerequisites

We expect to have proir knowledge of keras and tensorflow. Also see keras documentation for fitgenerator and ImageGenerator

You can use any version of the packages but we have used the following versions:

Packages	Versions
Keras	2.1.3
Tensorflow	1.8.0
Numpy	15.4.0
Matplotlib	3.0.2

Prior Work

We assume that your dataset is in the form of frames. These frames are located in every class-folder of training and validation. The frames of every instance are named with unique sequence numbers. Before getting to this point of work, your code might look like this

# Load entire dataset
from find_labels import findLabels
path_dataset_train='/home/activity/train'
path_dataset_val='/home/activity/val'

# Design model
model.Sequential()
.....
# Train model on your dataset
model.fit_generator()

Using DataGenerator Class

In python headers where packages are imported code this line

from DataGenerator import *
from dataset import load_cached

then pass path of the dataset folders to make or load cache_files for future usage

dataset_train=load_cached(cache_path='give path for making or loading .pkl file ',in_dir=path_dataset_train)
dataset_val=load_cached(cache_path='give path for making or loading .pkl file ',in_dir=path_dataset_val)
class_names=dataset_train.class_names

verify it by

print(class_names)

Now get images path and labels using

image_path_train,cls_train,labels_train=dataset_train.get_training_set()
image_path_val, cls_val,labels_val=dataset_val.get_training_set()
num_classes= dataset_train.num_classes

Concatenate both paths with parameters

partition={'train': image_path_train, 'validation': image_path_val}

Create and change parameters according to your needs for this 3D generators.

number_of_batches=10
frame_width=400
frame_height=256
frames_chunk=18
params={'dim': (frames_chunk,frame_width,frame_height),
        'n_channels': 3,
        'real_batchsize_custom': number_of_batches,
        'batch_size': frames_chunk*number_of_batches,
        'n_classes': num_classes,
        'shuffle': False
        'frames_chunk' : frames_chunk
        }

Get labels from the findLabels function

labels_training=findLabels(partition['train'], labels_train)
labels_validation=findLabels(partition['validation'], labels_val)

Pass the above created parameters to the main DataGenerator Class

training_generator = DataGenerator(partition['train'], labels_training, **params)
validation_generator = DataGenerator(partition['validation'], labels_validation, **params)

Give generators to your model.fit_generator() function for training the model

model.fit_generator(...
        generator=training_generator,
        validation_data=validation_generator,
        ..
        )

Authors

• Gulraiz Khan (Setting-up base of this project and generator with one batch)
• Abdulrehman Azam (Upgraded the generator for multiple batches)

License

This project is under GNU General Public License v3.0 see License file