This is a simple didactic demo that shows:
- how the image is processed all along the net (by showing the intermediate layer outputs) before returning the prediction results.
- the saliency maps of the input (ie. the maps that show which parts of the image where more influential in the decision making).
If you just want to make predictions as accurately as possible, use preferably deep.ifca.es/conus because, unlike this little demo, it uses random ten-crop batching and multimage averaging at test time to improve the predictions.
Submit image url
Submit local image