README_reproduce.md

Download the model

mkdir /export/share/projects/videogen/checkpoints/wildcamera/
cd /export/share/projects/videogen/checkpoints/wildcamera/
mkdir Release
cd Release
wget https://huggingface.co/datasets/Shengjie/WildCamera/resolve/main/checkpoint/wild_camera_all.pth?download=true
mv wild_camera_all.pth?download=true wild_camera_all.pth
wget https://huggingface.co/datasets/Shengjie/WildCamera/resolve/main/checkpoint/wild_camera_gsv.pth?download=true
mv wild_camera_gsv.pth?download=true wild_camera_gsv.pth
cd ..
mkdir swin_transformer
cd swin_transformer
wget https://huggingface.co/datasets/Shengjie/WildCamera/resolve/main/checkpoint/swin_large_patch4_window7_224_22k.pth?download=true
mv swin_large_patch4_window7_224_22k.pth?download=true swin_large_patch4_window7_224_22k.pth

Correct linking by going to the WildCamera code directory, then call:

ln -s /export/share/projects/videogen/checkpoints/wildcamera/ model_zoo

Download demo dataset

mkdir /export/share/projects/videogen/data/wildcamera
cd /export/share/projects/videogen/data/wildcamera
wget https://huggingface.co/datasets/Shengjie/WildCamera/resolve/main/asset/dollyzoom.tar
tar -xvf dollyzoom.tar
rm dollyzoom.tar
wget https://huggingface.co/datasets/Shengjie/WildCamera/resolve/main/asset/images-from-github-wt-intrinsic.tar
tar -xvf images-from-github-wt-intrinsic.tar
rm images-from-github-wt-intrinsic.tar

Correct linking by going to the WildCamera code directory, then call:

ln -s /export/share/projects/videogen/data/wildcamera data

Install additional dependencies. mmcv will take a while. mmcv is needed for inference.py, natsort is needed for dollyzoom,

pip install -U openmim
mim install mmcv
pip install natsort

Install current directory as a library.

pip install -e .

Test out demo scripts

# Estimate intrinsic over images collected from github
python demo/demo_inference.py

# Demo inference on dolly zoom videos
python demo/demo_dollyzoom.py

# Demo image restoration
python demo/demo_restoration.py