DeepChem has been great library for the application of deep learning for drug/chemical discovery. As of now there is no imaging models implemented in DeepChem that would facilitate the use of medical data (like images of brain scans, or UltraSound images etc). This project proposes to build an API for data augmentation for imaging which in recent times has shown to make models invariant slight data transformations like rotation, translation, noise etc. and as well as to build 2 models : one being the U-Net for bio-medical image segmentation and the other being the ResNet-50 model (trained on the imagenet), that would allow users/researchers to use these pre-trained models and extend it to implement their own networks with much ease.



Skand Vishwanath Peri


  • Bharath Ramsundar
  • Karl Leswing