Object detection is an AI that requires training with huge amount of dataset. Preparing the dataset is a trivial process as it has to be done manually. A lot of images has to be labelled manually using different tools such as Coco-Annotator and labelImg. Predict-and-Save helps the user to label new images using a pretrained model. This will reduce the workload of the users who want to prepare their own dataset.

predict-n-save.ipynb will generate dataset with 2 different formats, as different labeling tools use different dataset format.

1. COCO Json format
   Prediction will be stored in Json file with COCO format. With tools such as Coco-Annotator the user can import the json file and modify the annotations.
2. txt files with YOLO format
   Annotations are also saved in YOLO format. One common tools that uses this format to label images is labelImg.

Python 3.4, Tensorflow 1.3, Keras 2.0.8, numpy, scipy, Pillow, cython, matplotlib, scikit-image, opencv-python, h5py, imgaug and IPython

Modify categories.json with your own classes. In coco.py which is located in config folder, set the value of NUM_CLASSES with the number of your classes in categories.json.