kitti_playground

A playground to test sequences from the kitti dataset http://www.cvlibs.net/datasets/kitti/

Requirements

Download the tracking training sequences and labels from the kitti website or download a sample sequence from my google drive link. Note that Google Drive sometimes exhibits file preview issues, so do not download frames individually. Download the whole directory instead. If you want Mask-RCNN detections and masks you can download the compressed pickle results here

Libraries

• Opencv
• matplotlib
• numpy
• glob
• ntpath
• pickle
• gzip

Setting up virtualenv

virtualenv --system-site-packages kitti-env
source kitti-env/bin/activate
pip install -r requirements.txt

Add flag --system-site-packages to use TKinter for matplotlib.plot

Run the code

Run kitti_playground.py to see an example code

Usage

usage: kitti_playground.py [-h] labels_path images_path video_id

positional arguments:
  labels_path  Directory containing the labels, e.g. label_02/
  images_path  Directory containing the images, e.g. image_02/
  video_id     Video ID, e.g. 0000

optional arguments:
  -h, --help   show this help message and exit

Select video

Change video_id to match the names of the videos (0000, 0001, 0002, ...)

Select data

gen is a generator to obtain relevant data about each frame in the sequence. You can use loop=True to loop indefinitely over the video. You can pass to data_from_generator the types you want the generator to yield. Available types:

• 'img' -> the BGR frame
• 'annot' -> the ground truth data
• 'dets' -> Mask-RCNN detections
• 'masks' -> segmentation masks from Mask-RCNN
• 'feats' -> Mask-RCNN fc7 features for each detection box