There have been several GTA V mods released over the years that make the game more realistic, but none have even come close to photorealism.. Researchers Stephan R. Richter, Hassan Abu AlHaija and Vladlen Koltun might have the solution for that…machine learning. They managed to enhance the game by a convolutional network trained using a novel adversarial objective, which provides strong supervision at multiple perceptual levels.
The team analyzed scene layout distributions in commonly used datasets and find that they differ in important ways and then proposed a new strategy for sampling image patches during training. Multiple architectural improvements were also introduced in the deep network modules used for photorealism enhancement. When can we see something like this in the real world? Possibly in the very near future.
- Compact, lightweight handheld system
- Available in a range of fun colors
- Online and local wireless multiplayer
