diff --git a/README.md b/README.md
index 98b607a..40541fa 100644
--- a/README.md
+++ b/README.md
@@ -9,19 +9,37 @@ Code accompanying the paper ["Wasserstein GAN"](https://arxiv.org/abs/1701.07875
 - [PyTorch](http://pytorch.org)
 - For training, an NVIDIA GPU is strongly recommended for speed. CPU is supported but training is very slow.
 
+Two main empirical claims:
+
+###Generator sample quality correlates with discriminator loss
+
+![gensample](imgs/w_combined.png "sample quality correlates with discriminator loss")
+
+###Improved model stability
+
+![stability](imgs/compare_dcgan.png "stability")
+
 
 ##Reproducing LSUN experiments
 
 **With DCGAN:**
 
-```python
+```bash
 python main.py --dataset lsun --dataroot [lsun-train-folder] --cuda
 ```
 
 **With MLP:**
 
-```python
+```bash
 python main.py --mlp_G --ngf 512
 ```
 
+Generated samples will be in the `samples` folder.
+
+If you plot the value `-Loss_D`, then you can reproduce the curves from the paper. The curves from the paper (as mentioned in the paper) have a median filter applied to them:
+
+```python
+med_filtered_loss = scipy.signal.medfilt(-Loss_D, dtype='float64'), 101)
+```
+
 More improved README in the works.
diff --git a/imgs/compare_dcgan.png b/imgs/compare_dcgan.png
new file mode 100644
index 0000000..3b20913
Binary files /dev/null and b/imgs/compare_dcgan.png differ
diff --git a/imgs/w_combined.png b/imgs/w_combined.png
new file mode 100644
index 0000000..c17b8a1
Binary files /dev/null and b/imgs/w_combined.png differ