This project helps to semi-automate the labelling process for object detection model by converting the raw detections made by the pre-trained tensorflow object detection model to xml files in PASCAL VOC format. The xml files and the images can the be loaded into an annotation tool. The user then has to just tweak the bounding boxes and save making the process easier.
first clone this repository into your machine and then clone the tensorflow models directory into this directory, and follow the installation instructions for the object detection model given there. the models repository can be found here https://github.com/tensorflow/models
After setup, put the image you want to use into the input_images folder and run the main.py file with the image name as argument. Note that by default the image should be a .jpg file. however you can change in the main.py file.
python main.py --image_name
The output image with the detection overlay will be stored in the output_images folder and the xml file will be stored in the xml_files folder. And then you can take the images and their corresponding xml files and put into any annotation software. This was tested on labelimg tool and tensorflow 2.0