I’ve been working on an open-source tool to visualise attentions in Visual Question Answering models, and it’s finally ready!
I’m calling it AMA: Ask me anything.
Please check out the demo at apoorvesinghal.com/vqa
Step 1. Go to the website (preferably from a laptop/computer)
Step 2. Choose an image
Step 3. Ask a question
The system will try to answer any question you ask about the image!
Example: For the bottom left image, we ask the system “What is the colour of the hat?”.
The AI system answers: white.
The app also shows the parts of the image and the question that the system looks at, to answer the question!