Visual Question Answering using BLIP pre-trained model!
This implementation applies the BLIP pre-trained model to solve the icon domain task.
Description
**Note: The test dataset does not have labels. I evaluated the model via Kaggle competition and got 96% in accuracy manner. Obviously, you can use a partition of the training set as a testing set.
Create data folder
Copy all data following the example form You can download data here
Install requirements.txt
pip install -r requirements.txt
Run finetuning code
python finetuning.py
Run prediction
python predicting.py