Fast Depth Coding Using Deep Learning

Fast Depth Coding Using Deep Learning is the thesis topic. The details of the implementation are documented here for reference.

This documentation is organized into a couple sections:

• Propose the Algorithm
• Pre-process the Data
• Train the Model
• Evaluate the Model
• Use the Learned Model
• Simulation Results
• References

Propose the Algorithm

• Flow Chart
• Description
• [Deprecated]Flow chart

Pre-process the Data

• Data Collection
• Processing Steps and Statistics
• Data Visualization
• Edge Strength Analysis

Train the Model

• Convolutional Neural Network
• Settings
• Training for blocks of size 4x4
• Training for blocks of size 8x8
• Training for blocks of size 16x16
• Training for blocks of size 32x32
• Training for blocks of size 64x64

Evaluate the Model

• Model trained from blocks of size 08x08
• Model trained from blocks of size 16x16
• Using blocks of size 16x16
• Using blocks of size 32x32
• Using blocks of size 64x64

Use the Learned Model

• Time Cost of TF in C++
• Encoder Integration in C++

Simulation Results

• Simulation Results

Appendices

• About Compilation
• About GPU
• Thesis
• How to write thesis
• How to Obtain BD-BR & BD-PSNR
• Terms

References

• References