Caltech Pedestrian Dataset Converter to Pascal VOC 2007 format for Faster R-CNN

Requirements

• OpenCV 2.4.13
• Python 3.4+,
• NumPy 1.10+
• SciPy 0.16+

Usage (Caltech Pedestrian Dataset Example)

1. Prepare dataset

Download caltech dataset, pick up images and annotations from videos.

$ ./scripts/caltech-dataset-download.sh
$ ./caltech-parser/parse_annotations.py
$ ./caltech-parser/parse_seqs.py

Each .seq video is separated into .png images. Each image's filename is consisted of {set**}_{V***}_{frame_num}.png. According to the official site, set06~set10 are for test dataset, while the rest are for training dataset.

(Number of objects: 346621)

2. Draw and Test

You can draw bounding boxes in the images and get a video for checking.

$ ./scripts/test_plot_annotations.py

3. Create VOC2007 format dataset

Convert images to VOC2007 format dataset.

$ ./converter/converter.py --anno [annotations.json path] --images [images path] --dst [voc save path] --dataset [dataset type]

• config your properties for VOC2007 annotations and train and test sets in config.py.
• make voc_path dir empty is better.

4. Complete VOC2007 dataset

Copy other dir from VOC2007 like: local, results...

Notice

./converter/config.py is a configure file for different dataset. ./converter/filter.py can add yourself filter for different interesting objects.