Hello,
Given the shallow amount of data, it may be tricky to train a model avoiding overfitting. May we use any of the standard readily available datasets out in the community (Labelled Faces in the Wild, MORPH, etc.)?
Thanks!
Posted by: cristian.canton @ June 18, 2015, 4:49 p.m.Dear Cristian,
Absolutely. For age recognition challenge you can use any other data or pre-trained model you want (It is not the case of the cultural event recognition challenge, where you can only use the provided train and validation data).
Regarding my last post. To be precise, in the case of cultural event recognition: "External data regarding cultural events is not allowed, but is is allowed, in the case of deep models, to use models trained with other datasets instead of having to train the model from scratch. For instance, it is possible to use a network trained with ImageNet and fine-tune it to the problem at hand, but it's not allowed to use extra cultural images crawled from the internet."
However, for age recognition challenge you can use any data or pre-trained model.
Posted by: sergio.escalera.guerrero @ June 18, 2015, 5:22 p.m.Hi,
So we are allowed to use external dataset. We have our own in-house dataset that has not been published, can we use that as well? thanks.
Posted by: cpu90 @ June 27, 2015, 5:24 p.m.Dear Cristian,
Yes absolutely.
best regards