The task of this challenge is simple to describe, make a submission to Kaggle for the famous Dogs vs. Cats competition!
Here is a little bit extra requirements:
-
You MUST use Jupyter Notebook.
-
Not mandatory, but strongly recommend to use PyTorch.
-
In your notebook, include the public score of your submission to kaggle.
-
If your classifier is not 100% accurate, provide an analysis on the incorrectly classified inputs.
Take a harder challenge on Kaggle, Google Landmark Recognition Challenge. This dataset is huge, and it will take a long time to download, one of the participant is kind enough to provide an one-stop download of all the images in smaller size, see this discussion thread.
Besides training a high performance classifier, try to use visualization to understand which part of the image is the most salient to the classification decision (You may take a look into CAM: Class Activation Mapping).
Remarks:
-
If the data is no longer available at the time being, you may look for another Kaggle competition that interests you, build a model and try to visualize.
-
If you take this bonus part, the easy Dogs vs. Cats challenge can be skipped, but the above requirements are still applied.
-
A well done bonus part can outweigh everything else!