Now a days tensorflow 2.0 is taking over the world and object detection algorithms are the things in which researchers, students and indutries are intrested alot. So I decides to make a mask-rcnn compatible with tensorflow 2.0. I took help from alot blogs tensorflow 2.0 documentations and stack-overflow. Now a days its really complicated to made a tensorflow code which is compatible with nvidia 3000 series. So I also made some changes to make this code run on my 3070rtx. So tighten your seat belts, we are going to start our drive.

First step is to mannotate data I used VIA image annotator (https://www.robots.ox.ac.uk/~vgg/software/via/via-1.0.1.html) to annotate my data I make a rectangle against every object and save the project in json

After doing all the annotation things I rotate the image and box at every angle (0, 360) you can easily use my notebook to get done with your annotations.

After training Just opent he training notebook give the path of the parent folder where your augmented train and val located and start the training.

Now open the prediction notebook and just run the cells after specefiying weights paths and images path.