search

zhufengx / SRN_multilabel

144 stars 39 forks source link

Validation set included in training for WIDER #8

Open nsarafianos opened 6 years ago

nsarafianos commented 6 years ago

Thanks a lot for providing the code. I'm re-implementing your code on MXNet and I noticed that in the provided training scripts for the WIDER dataset you train on both the training and validation sets (with fixed steps for reducing the learning rate) and then test separately on testing so I'd like to ask you to make sure that this is the case.

Thank you

zhufengx commented 6 years ago

Hi, @nsarafianos , Yes, we use both "train" and "val" splits of WIDER dataset for training, and then test on "test" split. I also need to clarify that, in our provided training scripts the images in "**_val_imglist.txt" are randomly sampled from "test" split. This small portion of data is only used for viewing model status, but not for training.