I will be needing a step-by-step process for Image Segmentation or If you can break down the problem into comprehensible and programmatically implementable modules that would also be a great help.
Below Image is a snapshot of what needs to be done.
Here are some research papers that can help understand results I want.
- Automatic Portrait Segmentation for Image Stylization
- Automatically Remove Backgrounds From Images
- New Deep Learning Method Enhances Your Selfies
I have trained a tensor-flow model for custom object detection (apart from the 90 classes pre-trained COCO dataset model) but the problem is: TensorFlow does return the co-ordinates (bounding box) of the detected object, but we need the contour (outline) of the object.
^Considering that I know how to train CNN models with tensorflow, can you educate me how can I replicate the above research papers programatically? Like I know they collected a dataset of 1800 images and then transmuted that into 19000 images by augmentation but HOW can I prepare a dataset for training? what tool can I use to annotate images?
Please help me implement the above research papers. All I want is to extract a person from a photo - Portrait segmentation.
Thanks a lot!
