A CNN made using tensorflow that is cabable of detecting a pokemon on an image. The whole idea was to be able to identify pokemon from the hundreds of games available. The CNN architecture is based of a Mini VGG architecture.
The dataset for this project was made by scraping images of all first generation pokemon from a popluar pokemon website and then adding each image to various game/anime backgrounds. Finally a total of 1000+ images was collected for a 150 different pokemon in the first generation.
The scraper used to collect the images of pokemon is divided into 2 parts:
- scraper.py collects the pokemon and it's image url, saving it as a JSON file.
- downloader.py makes use of asynchronous functions to download each pokemon image
Finally the images (each pokemon and background image) is blended together
The CNN architecture is based of a Mini VGG architecture. It contains many of the same layers and operations used in state of the art CNNs today, however is a toned down version. For further understanding take a look at this research paper Activation Functions used are ReLU and softmax
- Add a CLI method to read image and predict the pokemon
- Add an API version of the same
- Add the next 2 generations of pokemon