Repository files navigation

Code and data for paper "Shading Annotations in the Wild".

Our code was tested on Ubuntu 14.04. As a first step, clone our repo:

git clone https://github.com/kovibalu/saw_release.git

Then install the python dependencies by running:

sudo ./install/install_python.sh

If you would like to run our trained model, you will need to installCaffe. We slightly modified the implementation ofBansal et. al for our purposes. To check out our Caffe version which is included as a submodule, run:

git submodule update --init --recursive

Then build Caffe after editing theMakefile.config depending on your configuration with:

cd caffemake all -jmake pycaffe -j

To download all data related to the dataset, run:

./download_saw.sh

The whole dataset download size is ~28.0GB, please see the documentation in thescript for a detailed breakdown of sizes for the different parts of thedataset. For detailed documentation on the format of the downloaded annotationsinsaw/saw_annotations_json seeANNO_FORMAT.md.

To generate the precision-recall curves in our paper for all baselines and our method, run:

python main.py generate_pr

You can select which baselines to evaluate inmain.py.

To generate the pixel labels from the SAW annotations andNYUv2 depth dataset depth and normal maps, run:

python main.py generate_labels

Please cite our paper if you use our code or data:

@article{kovacs17shading,author = "Balazs Kovacs and Sean Bell and Noah Snavely and Kavita Bala",title = "Shading Annotations in the Wild",journal = "Computer Vision and Pattern Recognition (CVPR)",year = "2017",}

Please contactBalazs Kovacs with any questions.