Object Detection and not using transfer learning

I recently read this well written article Practical -guide-object-detection-yolo-framework-python by Pulkit Sharma.
I wish to try Faster RCNN or YOLO frame work with out using their config files and pre-trained models and weights. I wish to start from the scratch like building the model, compiling the model, fitting the model with model .fit for training my data set(image, and xml file(bounding boxes). After running small number of epochs I wish to save the weights got by my custom model keras and then predict with model.predict. I appreciate your reply and links in this direction.

Hi @A.Malathi,

First of all as these are complex architectures, using the basic approach might not be possible as you have mentioned:

If you wish to train Faster RCNN model for your own dataset, refer to the following article:

1 Like

Can we implement person re-identification using the approach of object detection? If yes, then how we can proceed further and what will be the steps for that. I will use Market1501 dataset for person re-identification.

© Copyright 2013-2019 Analytics Vidhya