This repo is based on Pytorch tutorial for Image/video detection https://pytorch.org/tutorials/intermediate/torchvision_tutorial.html
In this repo I tried to fine-tune Mask RCNN for custom based dataset.
Here It's important to understand how to feed the data to the model
I have inherited the proprties of torch.utils.data.Dataset class
we need to overwrite init, call,getlen function.
In call if index of an image is given it should be able to read the image and
corresponding target and pass the image to transforms/augumentation before return.
Then dataloader will take care of the input pipeline , afterwards https://github.com/pytorch/vision/tree/master/references/detection
use the above link to get scripts which gives utility function for object detection.
Coco pre-trained weight can be obtained from https://pytorch.org/tutorials/intermediate/torchvision_tutorial.html
coco can be installed by using https://github.com/gautamchitnis/cocoapi.git@cocodataset-master#subdirectory=PythonAPI
install cython if make is failing