1. YOLO is a pre-trained model you can reuse for most of classes already added. if you want to get a list of the object please check the file coco.names which will have the list of 80 classes. Initially i tried recreating the yolo.h5 model by running yad2k.py over yolo.cfg and yolo.weights which i later figured was not required and reused pre-trained weights.
2. Its important to understand the concept of -
Encode Bounding boxes, Intersection over unoin and non-max suppression, anchor boxes.
Good place to refresh one's understanding -
3.As said earlier, though the YOLOv3 model should let you detect the object from the 80 classes mentioned, to detect any additional class you will need to custom train the YOLO model.
4.Finally though it may not be required to run the model, but to apply it to actual use cases one should be clear of the concept of CNNs. Few blogs i will suggest are -
https://victorzhou.com/blog/keras-cnn-tutorial/
https://victorzhou.com/blog/intro-to-cnns-part-1/
I tried the YOLO pre-trained weights and below are the results -
No comments:
Post a Comment