m3t4f1v3 / vbb2coco-converter

Download Caltech Pedestrian Dataset and convert them for Python users without using MATLAB

Caltech Pedestrian Dataset Converter

Requirements

• OpenCV 3.0+ (with Python binding)
• Python 2.7+, 3.4+, 3.5+
• NumPy 1.10+
• SciPy 0.16+

Caltech Pedestrian Dataset

$ sh download_and_convert.sh

Each .seq movie is separated into .jpg images. Each image's filename is consisted of {set**}_{V***}_{frame_num}.jpg. According to the official site, set06~set10 are for test dataset, while the rest are for training dataset. The .seq. is donwnloaded at ./data directory and the converted images are in ./data/images directory, divided into test and train folders. You can split your own test and train by modifying the ./config.yaml. The coco-format converted annotation is stored at ./data/annotations.

(Number of objects: 346621)

Draw Bounding Boxes

$ python tests/test_plot_annotations.py