Object Detection Using Faster RCNN

#1

I want to build a text localizer in images using Faster RCNN and not use pytesseract or OpenCV’s EAST detector but I am not able to understand the concept of variable output through the Neural network(varrying number of bounding box coordinates) and how their training and bachpropagation takes place as the number of output boxes is not fixed.
please help

#2

hey @Aryanchugh223 ,

There are very deep things to understand to get proper insight about this concept. You need to first understand about working of Faster RCNN , Region Proposal Networks , Convolutions and De-convolution and many different layers and most important understand how RCNN and FastRCNN works , if you just understand how they are working without going deep into there structure you can easily understand FasterRCNN.

#3

I did study about RCNN, FASTER RCNN AND YOLOs and how regional proposal katers work and ROI pooling from whatever I could find on web but still I am confused about some concepts and maths behind it so can you please suggest some good references for this

#4

if you want to understand the proper maths behind it , then let me tell you that its gonna take whole lot of time to get it done. I would recommend you to just understand the underlying concept ,working and not going into full depth about maths behind it .
You can have reference from these links:

  1. Faster RCNN for object detection
  2. Research Paper on Faster RCNN
  3. Understanding basic concepts
#5

Okay
Thankyou so much

#6

If there is something particular after these articles that is not understandable then you can surely ask me.

#7

I will surely do
Thanks for your help