1
$\begingroup$

I am getting into CNNs for several months already, and I am currently wondering myself if there is a difference between Retraining a DCNN and Fine Tuning it ?

I am working on a project in which my own new dataset is relatively small and very different from the original one. Then I don't really understand new what I have to look for.

$\endgroup$
1
$\begingroup$

Retraining a DCNN on another data set = Fine Tuning it

http://cs231n.github.io/transfer-learning/ (mirror):

Fine-tuning the ConvNet. The second strategy is to not only replace and retrain the classifier on top of the ConvNet on the new dataset, but to also fine-tune the weights of the pretrained network by continuing the backpropagation. It is possible to fine-tune all the layers of the ConvNet, or it’s possible to keep some of the earlier layers fixed (due to overfitting concerns) and only fine-tune some higher-level portion of the network. This is motivated by the observation that the earlier features of a ConvNet contain more generic features (e.g. edge detectors or color blob detectors) that should be useful to many tasks, but later layers of the ConvNet becomes progressively more specific to the details of the classes contained in the original dataset. In case of ImageNet for example, which contains many dog breeds, a significant portion of the representational power of the ConvNet may be devoted to features that are specific to differentiating between dog breeds.

$\endgroup$
1
  • $\begingroup$ Also, since this retraining modifies the last layer (the layer needed for the classification), would it be possible to turn this 'Classification' CNN into a 'Regression' one. For instance, getting the head position of dogs pictures. Would you be able then to retrain the DCNN using dog picture and (X,Y) coordinates as labels? And obviously modify/add few layers at the end. Or are the classification and regression cnn way too different ? Thank you again :) $\endgroup$ – HugoD May 12 '17 at 11:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.