Open pannNovember opened 3 years ago
Probably not! It also depends on the number of classes. So if you have 2000 datapoints for a binary (2 class) classification, it means 1000 points per each, so it might not be as bad. But let's say you have 10 classes, meaning that you have only 200 per class (which is probably terrible). If that is the case, then you could either look for knowledge transfer
articles or few shot learning
articles and use their tricks to make the network to work on your dataset. As a rule of thumb, having more data always helps.
This is for Cifar.If it is used with other dataset,what amount of data is required?Cifar is 50000 images.If other dataset is used,1000 or 2000 data ,is it possible to get good model?Does DenseNet121 require large data?Thanks!