KittiPy

Python functions to import and display the images and labels from the Kitty Object Detection Evaluation 2012 Dataset.

Prerequisites

• python ≥ v2.7.6
• scipy ≥ v0.19.1
• matplotlib ≥ v1.5.2
• numpy ≥ v1.13.3

Installing

1. Copy the file kittipy.py in the same directory as the python script or Jupyter notebook you want to use it with.
2. (Optional) Modify the kittipy.py global parameters at the begining
3. At the begining of your file, add the line: from kittipy import *
4. Enjoy!

How to use it

Explore Folder

To get the numpy array containing all the ids of the images contained in a folder, use the function:

get_data_list(im_set, db_absolute_path = ABSOLUTE_PATH)

Arguments	Default	Description
im_set		[string] 'train' or 'test' to chose the type of set
db_absolute_path	ABSOLUTE_PATH	[string] absolute path to the Kitti root folder

Import Image

To import images as a numpy array, use the function:

import_im(im_id, im_set, db_absolute_path = ABSOLUTE_PATH)

Arguments	Default	Description
im_id		[int] corresponding to the image id in the kitti dataset
im_set		[string] 'train' or 'test' to chose the type of set
db_absolute_path	ABSOLUTE_PATH	[string] absolute path to the Kitti root folder

Import Labels

To import the labels as a list of python dictionaries, use the function:

import_labels(im_id, im_set, db_absolute_path = ABSOLUTE_PATH)

Arguments	Default	Description
im_id		[int] corresponding to the image id in the kitti dataset
im_set		[string] 'train' or 'test' to chose the type of set
db_absolute_path	ABSOLUTE_PATH	[string] absolute path to the Kitti root folder

Description of the fields of the dictionary

Fields	Description
type	[string] 'Car', 'Van', 'Truck', 'Pedestrian', 'Person_sitting', 'Cyclist', 'Tram', 'Misc' or 'DontCare'
truncated	[float] Float from 0 (non-truncated) to 1 (truncated), where truncated refers to the object leaving the image boundaries
occluded	[int] Integer (0,1,2,3) indicating occlusion state: 0 = fully visible, 1 = partly occluded, 2 = largely occluded, 3 = unknown
alpha	[float] Observation angle of an object [-pi..pi]
bbox	[dict][float] x_min, y_min, x_max, y_max pixel coordinates (0-based index)
3D_dim	[dict][float] height, width, length (in meters)
3D_loc	[dict][float] x, y, z in camera coordinates (in meters)
rotation_y	[float] around Y-axis in camera coordinates [-pi..pi]
score (Optional)	[float] Only for results, indicating confidence in detection, needed for p/r curves, higher is better.

Display Image + Labels

To display a image and its labels in a Jupyter Notebook, use the function:

display_im(im, labels = [], display_boxes = True, display_info = True, 
           types_to_display = DEFAULT_TYPES_TO_DISPLAY, 
           info_to_display = DEFAULT_INFO_TO_DISPLAY, 
           db_absolute_path = ABSOLUTE_PATH, im_width = FIG_WIDTH, 
           im_height = FIG_HEIGHT, display_axis = False, 
           title = '', display_center_boxes = True, num_cell_grid = 0)

Arguments	Default	Description
im		[numpy.array] Image to display
labels	[]	[list][dict] List of dictionaries containing the labels of the image
display_boxes	True	[bool] Display the boxes around the objects
display_info	True	[bool] Write the info_to_display under the image
types_to_display	DEFAULT_TYPES_TO_DISPLAY	[list] List of types of object to consider
info_to_display	DEFAULT_INFO_TO_DISPLAY	[list] List of characteritics of object to display
im_width	FIG_WIDTH	[int] Width of the image to display
im_height	FIG_HEIGHT	[int] Height of the image to display
display_axis	False	[bool] Display the axis of the image
title	''	[string] Title to write before the image
display_center_boxes	True	[bool] Display the center of the boxes
num_cell_grid	0	[int] Display a num_cell_grid x num_cell_grid over the image

Remarks: To change FIG_FONT_SIZE_TITLE, COLOR_GRID and COLOR_TYPE do it in the global variables at the begining of the kittipy.py file.

Example

Code (Jupyter Notebook)

list_ids = get_data_list('train')
image = import_im(list_ids[3], 'train')
labels = import_labels(list_ids[3], 'train')
display_im(image, labels, num_cell_grid = 10)

Output

Author

• Lucas Vandroux (冯凯) 【Github】 【LinkedIn】

I started this project to help me during my Master Thesis at Tsinghua University in China. If you find it useful, don't hesitate to use it and improve it.

---