I trained a convolutional neural network with a large (>90k) dataset of images. Since there is some confusion between two classes, I collected all the images that were misclassified and I augmented them with several methods. Furthermore, I collected new images that were not present in the original dataset. Now I would like to re-train my network in order to reduce the confusion. What is the best approach? Should I perform fine-tuning using only the new images (case A) or should I add the new images to the whole training set and re-train the network from zero (case B)?

My concerns are the following: in case A, most of the images belong to the same class, therefore the dataset would be very unbalanced. In case B, I am afraid that the impact of the new images on the whole dataset would be very small.


1 Answer 1


I guess you can try both the solutions.

First try fine-tuning. Then, if the results aren't improving, retrain with all the data.

  • $\begingroup$ This is generally the best answer whenever a data science issues throws up "should I try model or training variation A or B". There is some small chance someone familiar enough with the problem has done similar enough work (similar data, models and problem), and they would have a clue. Otherwise, it is time to find out by experimentation. In this case, the OP has not given enough data for any expert to make a decent guess. $\endgroup$ Sep 24, 2018 at 16:19
  • 1
    $\begingroup$ You could detail a bit more in this answer what trying both solutions entails. The OP does need to be aware of using cross-validation so that they can compare approaches for instance, and it is not clear from the question that they are doing that. $\endgroup$ Sep 24, 2018 at 16:22
  • $\begingroup$ Thank you. I added some details in order to express my concerns better $\endgroup$
    – firion
    Sep 25, 2018 at 8:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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