Demo of the model deployed on the Ainize Platform.
This repository contains the docker file used to build the Svelte app shown in the GIF above. It uses a fine-tuned version of ViT trained on a dataset of Pokemon I collected. If you’re interested in using this model, I’ve also made it available on Hugging Face here.
- Data was collected using a mixture of Veekun images, PokemonDB, and images scraped from Brave Search.
- Training was performed using PyTorch-lightning and used ViT-base (see training folder for source code).
- Frontend demo was created using Svelte which was served using FastAPI.
If you'd like more information about this model, check out my Medium post.
The model was able to achieve 95% accuracy on my validation dataset. Though, feel free to check out the demo and try it yourself!