Do this if you need to create a UV parameterization for a new object class. Below the process is highlighted for the class car. For classes bird, car, horse, sheep, cow, zebra the uv parametrization exists in the cachedir
CSM_ROOT=/home/nileshk/csm_root/
cd $CSM_ROOT/csm/preprocess/meshdeform/
Please edit the startup.m
file with correct class_name
and output_map_file
.
output_map_file='/home/nileshk/CorrespNet/csm_root/csm/cachedir/downloaded_models/car/car_mapping.mat'
class_name='car'
Now create mapping from uv
to face ids.
python uv_to_vertex_map.py $CSM_ROOT/csm/cachedir/downloaded_models/car/car_mapping.mat $CSM_ROOT/csm/csm/cachedir/shapenet/car/shape.mat $CSM_ROOT/csm/csm/cachedir/downloaded_models/car/transform.mat
Follow instructions here
Install detectron using instruction here
We filtered images from ImageNet which were occuluded and truncated. Those annotations are in *_quality.json
file. Edit imageNetDir
and sysnetId
with the appropriate values.
cd $CSM_ROOT/csm/preprocess/imagenet/
python extract_class_data.py