Following tasks are done:
- Download the MS-COCO annotations and create a histogram of object classes.
- Next, extract two sets of 100 images in which the a) most often occurring object class is present, and b) least often occurring object class is present.
- For each of those images, replace the background (where no object is present) with a black-colored background.
- Find a tutorial to train a small GAN using PyTorch on the original sets of images and “ blacked-out” image sets.
- Report your findings.
I have referenced below mentioned online materials for completing the task (also mentioned in notebook):
- https://www.analyticsvidhya.com/blog/2019/07/computer-vision-implementing-mask-r-cnn-image-segmentation/
- https://towardsdatascience.com/master-the-coco-dataset-for-semantic-image-segmentation-part-1-of-2-732712631047
- https://towardsdatascience.com/master-the-coco-dataset-for-semantic-image-segmentation-part-2-of-2-c0d1f593096a
- https://pytorch.org/tutorials/beginner/dcgan_faces_tutorial.html