Own implementation of the most popular selfie algorithm with same neural network as in Google Pixel 3XL phone
Created using opencv and DeeplabV3+ with Xception71 backbone
run pip install -r requirements.txt
choose photo of arbitrary resolution and aspect ratio and save it in root project directory
substitute name of the file in line 7
run main.py file
- Eliminate bluring of random blobs that do not represent body shape
- Eliminate bluring parts of body as shoulder part in image 2
- Make transition from segmented body to blured background smoother
- Smooth detected body contour
- Try different segmentation nn architectures for example Unet