[opencv-gsoc-2020] Tutorial on training a network with TF or Pytorch

[Pragya Kapoor]

Hi, 

     I'm Pragya Kapoor, a third year B.tech Cse student at NIIT University. I'm interested in the project : Write a tutorial about training a network in TF or PyTorch and then running it in OpenCV. I have ample experience in this field and I have certain queries regarding this project : 

1. What is it that the mentors are looking forward to in this tutorial especially when it comes to choosing between TF and Pytorch. So, if the aim is to create great visualisations and have better deployment, etc then I think TF is a better option and we can even use Keras in this case but if the focus is on quickness, native feel(more pythonic) and dynamic over static then I think Pytorch is better. So, I would like an understanding of your liking over here.

2. What model do you suggest for this tutorial along with the choice of algorithm. Because I honestly feel that faster RCNN and YOLO are great choices for object detection specially for a live video feed. But if the focus is on accuracy then faster RCNN is better but if average accuracy but fast speed is required then YOLO is better. So, would like some clarity on this as well.

I have provided an in depth comparative analysis between TF and pyorch as well as between RCNN and YOLO in my draft proposal(PFA). Kindly have a look at it and I will be waiting for some feedback on the same.

Regards,

Pragya Kapoor

[Pragya Kapoor]

Hi,

    Hoping to hear from you soon.