This is a forked version of Amodal API. The original Amodal API is only compatible with Python 2. This version is compatible with Python 3 and Windows. This is the Python API code for the amodal segmentation dataset proposed in Semantic Amodal Segmentation (CVPR 2017). This API code is built on COCO API, which does not support Windows and the authors are not planning on supporting it in the future. However, there is a simple fix that works for both COCO API and Amodal API.

1. git clone and compile:

• git clone https://github.com/wakeupbuddy/amodalAPI
• cd PythonAPI; python setup.py build_ext install; cd ..
• Note that on Windows, an error will pop up if setup.py is run as is. This can be easily fixed by opening setup.py and removing extra_compile_args=['-Wno-cpp', '-Wno-unused-function', '-std=c99'],

2. create soft link for coco/bsds images:

• ln -s /your/coco/images ./images
• ln -s /your/bsds/images ./bsds_images

3. dowload annotation files and untar.

1. To see the annotation and some useful APIs, please run the ipython notebook demo.

1. dowload the baseline amodalMask output on coco val set and untar:

2. run the segmentation evaluation.

• bash eval.sh

It measures amodal segment proposal quality using average recall. Please see details in table 3a and section 5.1 from the paper.

We also release the web tool we used for annotation in another repo here. It's modified based on OpenSurface.

If you find this dataset useful to your research, please consider citing:

@inproceedings{zhu2017semantic,
    Author = {Zhu, Yan and Tian, Yuandong and Mexatas, Dimitris and Doll{\'a}r, Piotr},
    Title = {Semantic Amodal Segmentation},
    Booktitle = {Conference on Computer Vision and Pattern Recognition ({CVPR})},
    Year = {2017}
}