Update README.md

README.md

@@ -4,6 +4,10 @@ COFW sample images segmented using our method.
 
 [Yuval Nirkin](http://www.nirkin.com/), [Iacopo Masi](http://www-bcf.usc.edu/~iacopoma/), [Anh Tuan Tran](https://sites.google.com/site/anhttranusc/), [Tal Hassner](http://www.openu.ac.il/home/hassner/), and [Gerard Medioni](http://iris.usc.edu/people/medioni/index.html).
 
+## News (10/07/18)
+- New FCN model released for lower resolution images (300X300), trained without augmentations. Useful if you have limited GPU memory.
+- A better performing and more efficient U-Net model will be released soon, including training and inference scripts using PyTorch.  
+
 ## Overview
 This project provides an interface for face segmentation using Caffe with a fully convolutional neural network.
 The network was trained on IARPA Janus CS2 dataset (excluding subjects that are also in [LFW](http://vis-www.cs.umass.edu/lfw/)) using a novel process for collecting ground truth face segmentations, involving our tool for [semi-supervised Face video segmentation](https://github.com/YuvalNirkin/face_video_segment). Additional synthetic images were generated by augmenting hands from the [EgoHands dataset](http://vision.soic.indiana.edu/projects/egohands/), and augmenting 3D models of glasses and microphones.