Stitching Images to create one seamless panorama image using Homography.
The purpose of this project is to stitch two or more images in order to create one seamless panorama image by finding the Homography between the two images. The project is divided into two phases,
- Phase 1: Classical approach of local feature matching
- Phase 2: Deep Learning approach(Homography Net - supervised and unsupervised) to estimate the homography.
- Mandeep Singh
- Chinmay Kate
Implemented traditional CV pipeline combines algorithms of corner detection, ANMS, feature extraction, feature matching, RANSAC, homography estimation and blending.
-
Open directory Phase1/Code and run the following command with the Data location as command line argument: -
python3 Wrapper.py --DataPath Set1
Above command will load Set1 images which needs to be stitched.
-
Results
folder contains stitched images of all Trainsets and TestSets.
In Deep learning, used Homography Net (both supervised and unsupervised) to estimate the homography.
To generate dataset, run the following command in Phase2/Code/supervised: -
python3 Wrapper.py
-
To train the network, run: -
python3 Train.py
-
To test the model on test set, run: -
python3 Test.py
-
To train the network, run: -
python3 Train.py
-
To test the model on trainset , run: -
python3 Test.py