Image compression

starter code

In this project we will implement a image compression network. The format will be similar to project 1, where you are expected implement a encode and a decode function. Your encoder/decoder are expected to perform three levels of compression, compressing 256x256 images to latent vectors no larger than 4kB, 16kB, and 64kB respectively. You may use any architecture to achieve this task, but using one of the following algorithms:

• Full Resolution Image Compression with Recurrent Neural Networks
• Real-Time Adaptive Image Compression
• Soft-to-Hard Vector Quantization for End-to-End Learning Compressible Representations
• End-to-end Optimized Image Compression

We provide a python starter code which contains a project and a val_grader. During testing we will evaluate your algorithm on the image quality on the reconstructed images on all three compression levels:

Evaluation:

• L1 distance
• SSIM
• Perceptual Loss

You can use python -m val_grader project -v to test your solution against the grader.