As in 1st class the model had really difficulty with differing some breeds like ‘Egyptian_Mau’, ‘Bengal’.
In real world setting when that’s is a problem and you want more accurate results what would you do?
Get more data for the trouble some labels or are there any special trics like TTA?
1 Like
You can try building a separate and subsequent model just for those classes, and that model with perhaps a different architecture and different transforms.
1 Like
We’ll be spending much of the rest of the course studying this!
3 Likes