roipoly-improved: Image ROI Labeling

How to launch

python3 image_labeling.py --roi-type stopsign,other_red,green,blue,yellow,gray,white [path to your dataset]

Before launching the program, make sure the train-set folder doesn't contain images you don't want to label

More detailed command

usage: image_labeling.py [-h] [--roi-type ROI_TYPES] [--output PICKLE_NAME] image_dir

Label image ROIs

positional arguments:
  image_dir             the directory of images to be labelled

optional arguments:
  -h, --help            show this help message and exit
  --roi-type ROI_TYPES  preload some roi types, separated by ","
  --output PICKLE_NAME  the name of pickle file contains mask

How to label

Do as the following figure shows.

If you don't hit 'Finish' before closing the windows, the current labeling result won't be saved!

You can review the mask meanwhile. This mask is just for visual check.

How to load result

By default, this tool saves the result to mask.pickle. Here is a sample Python3 code for loading the pickle file:

import pickle
with open('mask.pickle', 'rb') as f:
    a = pickle.load(f)

Sample output:

>>> a[0]  # image mask 
{'99.jpg': array([[0., 0., 0., ..., 0., 0., 0.],
       [0., 0., 0., ..., 0., 0., 0.],
       [0., 0., 0., ..., 0., 0., 0.],
       ...,
       [0., 0., 0., ..., 0., 0., 0.],
       [0., 0., 0., ..., 0., 0., 0.],
       [0., 0., 0., ..., 0., 0., 0.]])}
>>> a[1]  # roi types
{1: 'stopsign', 2: 'other_red', 3: 'green', 4: 'blue', 5: 'yellow', 6: 'gray'}

To extract the roi of specific type you want, here is a sample code:

import numpy as np
import matplotlib.pyplot as plt
mask = a[0]['99.jpg']
roi = plt.imread('99.jpg')[mask == 1]  # get stop sign roi

Attention

Because this tool use 0 to mark unlabeled pixel, the minimum label index is 1.

Other Tools

pickle_merge

If you have multiple pickle files created by image_labeling.py, we can use pickle_merge to merge these files into a single one.

It also convert mask array to np.array(dtype=np.uint8), which reduces 75% file size if those pickles are stored as int32.

Here is an example of using pickle_merge.py

python3 pickle_merge.py mask1.pickle mask2.pickle