Please use Python 3.6, opencv-contrib-python (3.4.0.12) and Pytorch (>= 1.1.0).
We use DEGENSAC from https://github.com/ducha-aiki/pyransac
Other dependencies should be easily installed through pip or conda.
For a quick start, set the dataset_path in demo/run_bash.sh as the path for test data, run bash run_ransac.sh. A dump folder named 'fundamental' will be created with corresponding inlier corr files and estimated F. For the estimation of E, please change the dump_path, model_path in run_ransac.sh and set use_fundamental=false, this requires a 'K1_K2.h5' file for every sequence.
There are five files for each test sequence,
'score.h5': Prediceted confidence score for each correspondence.
'corr_th.h5': Inlier correspondences by applying a thresholding(>1 by default) on the aforementioned confidence score.
'E/F_weighted': E/F estimated with weighted 8-point algorithm.
'corr_post': Inlier correspondence surviving from score thresholding and degensac.
'E/F_post': E/F estimated with degensac using corr_th.
In most cases, 'E/F_post' are the most precise.
For the parsing and evaluation, please refer to demo/eval_ef.py