micco00x / py-pascalpart

Py-PascalPart is a simple tool to read annotations files from PASCAL-Part Dataset in Python. It has been developed as final project for the module Human-Objects Relations of Elective in AI (Spring 2018) at Sapienza University of Rome.

Py-PascalPart

Py-PascalPart is a simple tool to read annotations files from PASCAL-Part Dataset in Python, giving an easy access to 10103 images, 24971 object masks and 181770 body parts.

Repository and Datasets

Clone the repository and download the PASCAL-VOC 2010 and the PASCAL-Part Datasets if you don't have them on your computer:

git clone https://github.com/micco00x/py-pascalpart
mkdir datasets
cd datasets
wget http://www.stat.ucla.edu/~xianjie.chen/pascal_part_dataset/trainval.tar.gz
wget http://host.robots.ox.ac.uk/pascal/VOC/voc2010/VOCtrainval_03-May-2010.tar
tar -zxvf trainval.tar.gz
tar -zxvf VOCtrainval_03-May-2010.tar
rm VOCtrainval_03-May-2010.tar
rm trainval.tar.gz

Explore the dataset:

python3 explore_dataset.py

Explore the dataset specifying a different path for PASCAL-Part Dataset annotation folder and PASCAL VOC 2010 JPEG images folder:

python3 explore_dataset.py --annotation_folder=PATH_TO_ANNOTATIONS --images_folder=PATH_TO_IMAGES

Note that utils.load_annotations(path) returns a dictionary containing all the information stored in the .mat file specified by path. See explore_dataset.py for a typical example of usage. More in detail, once the annotation file has been loaded:

• annotations["objects"] is a list of objects contained in the image
• obj["mask"] is the mask of the object
• obj["class"] is the class of the object
• obj["parts"] is a list of body parts
• body_part["mask"]is the mask of the body part
• body_part["part_name"] is the class of the body part

with obj element of annotations["objects"] and body_part element of obj["parts"].

py-pascalpart's Issues

COCO transform

Hello,

Very nice and useful repo! Are you aware if there is any code to transform these annotations to COCO format?

error: Error -3 while decompressing data: invalid code lengths set

I keep getting the following error: "error: Error -3 while decompressing data: invalid code lengths set". In the following rule: annotations = scipy.io.loadmat(path)["anno"]. Does anyone else had this issue? Or knows how to solve. Thanks in advance