These scripts are used for convert datasets (MS COCO, Caltech pedestrian dataset) to PASCAL VOC format for later training.

Requirements

• MS COCO toolbox
• cytoolz
• lmxl
• scipy, numpy

Usage

anno_json_image_urls.py: extract image url (coco source not filckr) from annotation json file. See anno_json_image_urls.sh
download_coco_images.py: download coco image files from given urls (extracted from instance/keypoint annotation json file) . See download.sh
anno_coco2voc.py: convert coco annotation json file to VOC xml files. See anno_coco2voc.sh
vbb2voc.py: extract images with person bbox in seq file and convert vbb annotation file to xml files.