Closed asanakoy closed 4 years ago
FastAP generally works well with few-shot learning datasets but there are no limitations to what types of datasets that can be used.
Just as a caution, the mini-batches should be constructed in a way similar to the proposed sampling method in the paper.
Hi,
Nice work! Do you guys have any quantitative results for CUB200-2011 and CARS196 datasets? Are there any limitations which doesn't allow to train FastAP on such small datasets?