- Prepare GAN models(Faces, MNIST, Rooms) for different generator priors(Normal, Uniform)
- Implement interpolation functions from the paper:
- 2-point interpolation
- n-point interpolation
- vicinity sampling
- analogies
- Conduct experiments and compare results
- Prepare VAE models(Faces, MNIST, Rooms)
- Implement functions from the paper:
- 2-point interpolation
- n-point interpolation
- vicinity sampling
- analogies
- Implement new interpolation for latent space of VAE
- Conduct experiments and compare results
The idea is to check whether it’s possible to combine 2 pictures with missing parts into 1 good image by first mapping into latent space, performing interpolation and then mapping back using decoder(Using different interpolation techniques). So, for this we will need encoder-decoder architecture. We are planning to use VAE for the moment.
Draw 2 dimensional map of the dataset using vicinity sampling