Yolo is trained from scratch on Vooc 2007 dataset: http://host.robots.ox.ac.uk/pascal/VOC/voc2007/
Basic architecuture remains same only the major difference is that we need the last layer to predict a grid matrix of shape 7X7X30 where last dimension is composed of 20 classes and two bounding boxes with object probablities.
The two main function are Yolo_reshape and yolo_loss. First one is to create customized last layer and second one is for customized loss function.
Note: The training was taking a lot of time. Need to update again once the model training is complete