Hi,
The competition rules do not specifically prohibit human-assisted approaches, so I ask if any of the following schemes is allowed:
1. Can the participants manually crop (or correct an incorrect crop of) the facial area in each picture (also in the test set released later)?
2. Can the participants manually add landmarks to the facial areas (e.g., nose, ears, eyes)?
3. Can the participants manually guess the age (to reach 0.34 score)?
Obviously item 3 is not allowed (right?), but I'm curious about the first two. Many age estimation approaches report the accuracy using human entered landmarks, and a manual annotation of the yet unreleased test set should not take more than a few hours / days.
Heikki
Dear Heikki,
Option 3 is not allowed.
Options 1 and 2 are allowed with some details:
Manual crop and addition of landmarks are allowed only on the training set or any other external data but NOT allowed in the validation set during development phase.
For final test set when labels are provided for the validation set you can also crop and edit landmarks on the validation set but NOT allowed for the test set. I mean, the data are used for testing at each step CANNOT be manipulated manually, just used as input for testing your whole method. Testing data at each step has to be completely automatically tested. The data considered for training can be manipulated as you want in order to help training your approach.
best regards