TAO Classification - Unbalanced Dataset

I am training a classification network (efficientnet_b0) on a dataset where I have unbalanced images.

Does each minibatch have equal number of examples from each class of data? Are there any config options for how the data is loaded or how sampling is handled?

• Hardware: V100
• Network Type: Classification

No. It is not needed.

There is not parameters in the spec file.

Thanks Morganh - any links or references on why minibatches aren’t balanced? Is some other technique used (weights or something)?

Thanks for the info on not having data set options via the spec file.

Sorry, each minibatch has equal number of examples from each class of data.
I misunderstood your question. What I mentioned earlier is that for training images, it is not needed to set equal number of examples for each class.

There is no update from you for a period, assuming this is not an issue anymore.
Hence we are closing this topic. If need further support, please open a new one.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.