Synthetic Depth of Field with and without Depth using PyTorch

Introduction

This respository contains code for applying synthetic depth of field to images. Synthetic depth of field is common in modern smartphones, for example, portrait mode images in iPhones and Pixel phones, and live focus mode in Samsung phones. If depth information is not available then the code uses MiDaS. The code is pretty simple.

Requirements

• PyTorch
• NumPy
• OpenCV
• Matplotlib

How to use

The parameters for the code are given in the options.py file. Basic use without depth is as follows.

python3 main.py --im_name test_ims/rocks.jpg

The depth map is saved in the depth_dir.

For synthetic depth of field with input image and depthmap from Pixel 4 (or any other depth map), usage is as

python3 main.py --im_name test_ims/eraser_input.jpg --depth_name test_ims/eraser_depth.png 

The result is stored in output_dir.

Some Results

Pixel 4 result

Random image result