Skip to content

Latest commit

 

History

History
37 lines (28 loc) · 1.34 KB

README.md

File metadata and controls

37 lines (28 loc) · 1.34 KB
Raw

Patch Match

The algorithm of Patch Match implemented in Python. Adapted from this repo, it is a method for warping the satellite labels to match the drone images.

1.Introduction

This code mainly implements the algorithm of patch match, for details, please see the paper: PatchMatch: A Randomized Correspondence Algorithm for Structural Image Editing.

2.Setup

Run the following commands:

pip install numpy tqdm numba Pillow matplotlib opencv-python

3.Run

Example of command for generating our warped labels:

python PatchMatch.py --data_dir ../DroneVision/drone_dataset/data/ 
                    --sat_dir img_dir 
                    --drone_dir drone_dir 
                    --label_dir ann_dir
                    --output_dir patchmatch_warped
                    --p_size 3
                    --itr 5
                    --new_w 1056
                    --new_h 792
                    --use_my_init
                    --img_type 'drone'
                    --ref_type 'sat'

The first 5 parameters are input/output related and the next are Patch Match algorithm's parameters. You can find further explanation of the parameters in the get_cli_arg() function in PatchMatch.py script.