This is a Keras implementation of a Light Weight CNN for Trimap Prediction. I trained Densenet based trimap prediction model. Natural image matting is an important problem in computer vision and graphics. It is an ill-posed problem when only an input image is available without any external information.In practice, most existing matting methods take a trimap as input; however, matting is still underconstrained in the undefined area in the trimap. I created dataset and model which predict the trimap for different of kind of objects.
- Python3.6+
- Windows 11, Python 3.6.1, Tensorflow 1.10.0, CUDA 10.0
Context-Aware Image Matting for Simultaneous Foreground and Alpha Estimation