Clone the repository:
git clone --recurse-submodules git@github.com:AyonRRahman/Thesis.git
Download the Eiffel Tower dataset and use colmap to undistort images
cd data
chmod +x datasetup_Eiffel_Tower.sh
./datasetup_Eiffel_Tower.sh
Prepare the dataset for training in SC-SFMlearner:
run the crop_and_downscale_colmap_undistorted_image() function in utils.data.py to prepare the the Eiffel-Tower dataset for training.
Create the ground truth trajectory using utils.utils.export_trajectory() function in the kitty format.
Or undistort using opencv to keep the image dimension same and prepare for training:
cd data
python undistort_Eiffel_tower_using_opencv.py
use Downscale_image function from utils.data to downscale the images.
Using these repositories:
Maxime Ferrera's Script and OpenMVS.
Build Maxime Ferrera's Script according to the mentioned method in the repository in the thirdparty folder.
Use docker file given in the OpenMVS to build a docker container.
convert the colmap model and start the docker for openMVS in the undistorted(colmap) folder
python create_txt_files_for_colmap_model.py
../Thirdparty/openMVS/docker/QUICK_START.sh ./Eiffel-Tower_undistorted
In the terminal of the docker container for each folder(2020, 2015, 2016, 2018) of data do this:
InterfaceCOLMAP -i ./2020 -o ./2020/scene.mvs --image-folder ./2020/images
ReconstructMesh ./2020/scene.mvs -o ./2020/scene.ply
now from the native terminal:
sudo ./Create_depth_images.sh
use the functions crop_and_downscale_colmap_undistorted_image to crop and downscale images and depth. Use generate_mask_from_depth to get the masks.
Official link to download the dataset: http://www.cvlibs.net/download.php?file=raw_data_downloader.zip
cd data
mkdir Kitty_raw
cp Kitty_raw_data_downloader.sh Kitty_raw
cd Kitty_raw
./Kitty_raw_data_downloader.sh