![]() This was to account for the disparity of walls to players in the training set. When saving a dataset, this flag gets toggled after every image. ![]() If the variable multiplayerMode=true, the generator will cram a lot of player triangles into the image instead of walls. This will generate one image and csv of bounding boxes per frame and save it to "generated/". To actually save a full dataset, you'll need to open "SaveImages.pde" and update line 1 to set saveImages to true (and maybe numImages to set your dataset size). r - randomize params for generated image.b - toggle on/off generated image bounding boxes in polar mode.Left Mouse - toggle between real images (from "img" directory) and generated image.I gave myself a few inputs for debugging purposes: Located in the "HexagonGenerator" directory is my Processing sketch that generates images that look like screenshots from Super Hexagon, with objects on screen labeled with bounding boxes. It's not as cleaned up as I would like and I am making no guarantees that it is in a working state. I am releasing this code as is, more than a year after I last worked on it. The project is described in detail in this report and demonstrated in these videos playing levels 1 and 2: the Python code that plays Super Hexagon using the trained CNN.the Python code that trains the CNN using Theano/Lasagne.a Processing sketch that generates imitation screenshots of Super Hexagon. ![]() This is my code for my CS231n project at Stanford for a convolutional neural network (CNN) that can play the game Super Hexagon. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |