PASCAL Visual Object Classes (VOC) Challenge: The challenge is a benchmark in visual object detection, it provide a standard dataset of images and annotation, which has been used in Faster RCNN and YOLO.
read_xml.ipynb: In this notebook, I write a script to read xml file and draw a bound box on the image.
Figure 1. An example of labeled images. The bound box information inside folder annotations.
To achieve the automatic detection of light echoes, the first thing needed is to build a image dataset with high quality annotation. I label some image in the format of PASCAL VOC. Next I will try to use these image to train a model and test the performance of different models.
Ref.
PASCAL VOC challenge http://host.robots.ox.ac.uk/pascal/VOC/pubs/everingham10.pdf
Pascal dataset mirror
https://pjreddie.com/projects/pascal-voc-dataset-mirror/
Image label tools:
-
LabelImg
-
Images Annotation Programme https://github.com/frederictost/images_annotation_programme/
-
RectLabel (for Mac )