Updated readme

README.md

@@ -15,14 +15,14 @@ datasets:
 
 MADNet is a deep stereo depth estimation model. Its key defining features are:
  1. It has a light-weight architecture which means it has low latency.
- 2. It supports self-supervised training, so it can be convieniently adapted in the field with no training data. 
- 3. Its a stereo depth model, which means its capable of much higher accuracy than mono depth techniques.
+ 2. It supports self-supervised training, so it can be conveniently adapted in the field with no training data. 
+ 3. It's a stereo depth model, which means it's capable of high accuracy.
  
  The MADNet weights in this repository were trained using a Tensorflow 2 / Keras implementation of the original code. The model was created using the Keras Functional API, which enables the following features:
- 1. Good optimization out the box. 
+ 1. Good optimization. 
  2. High level Keras methods (.fit, .predict and .evaluate).
- 3. Less boilerplate code.
- 4. Decent support from external packeges (like Weights and Biases). 
+ 3. Little boilerplate code.
+ 4. Decent support from external packages (like Weights and Biases). 
  5. Callbacks.
  
  The weights provided were either trained on the 2012 / 2015 kitti stereo dataset or flyingthings-3d dataset. The weights of the pretrained models from the original paper (tf1_conversion_kitti.h5 and tf1_conversion_synthetic.h5) are provided in tensorflow 2 format. The TF1 weights help speed up fine-tuning, but its recommended to use either synthetic.h5 (trained on flyingthings-3d) or kitti.h5 (trained on 2012 and 2015 kitti stereo datasets).