VQA Browser is web browser based tool for evaluating and visualizing your algorithm on VQA dataset Once you build a VQA result file of your model, getting quantitative evaluation and browsing the model's prediction resutls are much easier with VQA Browser.
- Qualitative Evaluation
- Browsing qualitative results
- Visualizing additional results (e.g. attention map)
- Searching questions
- Quantitative Evaluation & Comparison
- Computing VQA score on validation set
- Score comparison of multiple models
- Plotting comparison results
- Merging two model's results (picking the best answer for each question)
Clone our repository.
git clone https://github.com/HyeonwooNoh/VQABrowser.git
Run the following script.
./setup.sh
This script will download MSCOCO images and create simulink. If you have MSCOCO images in your local machine already, you can simply create a simulink.
Launch the server
python browser.py --host [host] --port [port]
Open http://[host]:[port] with your web browser to interface the VQA Browser.
