Closed ClancyAyres closed 1 year ago
I think you may on an old code branch, please ensure your code is consistent with the newest main branch.
i have the same problem, when i eval the thumos dataset, it work nomally. but when i eval the anet, some warning occured to it.
My understanding is: the number of training is too small, and some categories have not yet appeared. If the parameters after multiple rounds of eval, this problem will disappear
This can happen when evaluating an early checkpoint of the model, as the concepts of those categories have yet been learned.
For ActivityNet, please make sure that the same config is used for training and evaluation. A similar error can occur if the external score is enabled during training yet disabled at inference time.
Closed due to inactivity.