Pokedex using CNN (Image Recognition) and RAG
Link to open source: https://github.com/Wriath18/PokeDex_Streamlit_LLM
- Upload an image of a Pokémon to classify it using the CNN model.
- Automatically embed documents related to the Pokémon for detailed information.
- Generate and display a description of the classified Pokémon.
- Provide an input box for users to ask further questions about the Pokémon.
-
Pokemon Recognition : The pokemons are recognized using a CNN (Convolution Neural Network) model which has been trained on a Pokemon Image Dataset of 10 classes (10 Pokemons), the model has been trained on Kaggle using P100 GPU and the notebook can be accessed through the Model Training Folder.
-
RAG : The model generates descriptions and replies to question and answers by implementing RAG (Retrieval Augmented Generation) on a pdf file which contains information of the 10 pokemons "pokemon_info.pdf"
-
LLM Model : The LLM model used is Llama-3 8b model and is being inferenced using Groq API for faster working.
-
DB - For the RAG vector db has been used to create document embeddings.
