Skip to content

Create all types of object detection models with their backbones

Notifications You must be signed in to change notification settings

mitran27/Object_Detection

Repository files navigation

Object_Detection

Create various object detection models and backbones Entirely from scratch using pytorch /tensorflow like Tradional,Rcnn,faster Rcnn, SSD,Fpn,RetinaNet ,RetinaFace

After building state of the models in classification object detection model started to evolve

Traditional detection

Traditional object detection uses the concept of Sliding Window and pretrained classification model classifies the cropped image

RCNN

Rcnn uses the algorithmSelective search where initial proposals are created using felzenszwalb segmentation and all the proposed region are given to pretrained classifier

Fast Rcnn

selective search is used as region proposals , the roi pooling is done on region proposals correspondingly in the feature maps from the backbone network and the roi's are passed to twin network of object class and bounding box regression to detect the object

Faster Rcnn

RPN is used as region propasal which uses twin network on the feature maps created with the shared backbone of Faster rcnn rpn is trained with the help of anchors for proving region of anchors anchors are default boxes with different size and scales present in different regions of the input image rpn need to find the anchors which has possibility of finding the object and precise location of the object with respect to anchor coordinated

About

Create all types of object detection models with their backbones

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published