Goluck-Konuko / hevc_data_augmenter

Data augmentation module for training neural networks to perform HEVC intraprediction

On-the-Fly Data Augmentation for Training Neural Networks to Perform HEVC Intra Prediction

This module uses the standard HEVC implementation for Intra-DC, Planar, and 33 angular directions.

Data Preparation

The Augmenter class allows for flexible data input methods:

• filepath: Specify the file path when instantiating the augmenter object.
• odp: Supply an ODP (Original Decoded Picture) as a NumPy array.
• block_size: Specify the block size.
• bit_depth: Specify the bit depth.

Example:

from hevc_augmenter import Augmenter 
augmenter = Augmenter(odp=[optional], filepath=[optional], diskpath=[optional], block_size=[optional], bit_depth=[optional]) 

Usage

a) Check the Acquired Reference Samples

left, top, context, original_pu = augmenter.read_context()
print('Left reference samples:', left)
print('Top reference samples:', top)

b) Check the Result of Reference Sample Interpolation

left, top = augmenter.interpolation()
print('Left reference samples:', left)
print('Top reference samples:', top)

c) Check Reference Samples After Linear Filtering

left, top = augmenter.filter_reference_array()
print('Left reference samples:', left)
print('Top reference samples:', top)

d) DC Prediction

prediction = augmenter.intra_prediction_dc() print(prediction)

e) Planar Prediction
prediction = augmenter.intra_prediction_planar()
print(prediction)

f) Angular Prediction
prediction = augmenter.predict_one(mode=[optional])
print(prediction)

g) Run all prediction modes
prediction = augmenter.predict_all()
print(prediction)