Update README.md

.gitignore

@@ -242,4 +242,8 @@ paket-files/
 
 # JetBrains Rider
 .idea/
-*.sln.iml
+*.sln.iml
+
+# Custom
+media/
+logs/

README.md

@@ -1,9 +1,17 @@
-# tfworldhackathon
+# MusicGAN
-GitHub repo for my Tensorflow World hackathon submission
+**Scored in the Top 10 most innovative artificial intelligence projects in the international TFWorld Hackathon 2020.**
 
-[Video demo](https://www.youtube.com/watch?time_continue=2&v=DQDYsJXLqlo&feature=emb_logo)
+MusicGAN creates 1 second of instrumental audio at 16kHz.
 
-To replicate my results, just clone this repo and run `scripts/WGAN-GP.py` after modifying `DATA_DIR`. I used [this](http://deepyeti.ucsd.edu/cdonahue/wavegan/data/mancini_piano.tar.gz) data after converting the files to wave.
+[Piano demo](https://www.youtube.com/watch?time_continue=2&v=DQDYsJXLqlo&feature=emb_logo)
+
+[Violin demo](https://www.youtube.com/watch?v=XcESDdsb3PU)
+
+To create your own MusicGAN, just clone this repo and run `scripts/WGAN-GP.py` after modifying `DATA_DIR` and `INSTRUMENT`. I used [this](http://deepyeti.ucsd.edu/cdonahue/wavegan/data/mancini_piano.tar.gz) data after converting the files to wave. The violin data I used was just scraped from Youtube.
+
+Training expects a GPU and will take several hours to achieve resonable results. Due to time limitations for the hackathon, I only trained for a few hours, but better results are possible if trained for a longer duration.
+
+Run `tensorboard --logdir logs/train` to view generator and critic loss with Tensorboard.
 
 ## Inspiration
 Since the inception of generative adversarial networks, I have been fascinated by their capacity to perform tasks of unprecedented complexity. They are a prime example of how machines can learn in a similar manner to humans - akin to reinforcement learning. I am also a huge fan of music and love to play the piano. So I thought, why not conflate my love for machine learning and my passion for music!?