I find applications of convolutional networks and sequence models especially interesting.
One of the ideas I have is to build a command-line tool to classify my photos (mark photos with people, nature, cities, similar photos, photos with bad light / focus, ...).
Could you suggest progression through such project with high-level tasks ordered by difficulty?
Other project suggestions are also welcome!
1. establish desired categories
2. group existing photos into these categories
3. finetune a pretrained model on these categories
You can also partially automate step 2 using a pretrained model, but in this case some of your desired classes should be similar to (some of) the classes that model was trained on.