YoungXIAO13 / FewShotDetection

(ECCV 2020) PyTorch implementation of paper "Few-Shot Object Detection and Viewpoint Estimation for Objects in the Wild"
http://imagine.enpc.fr/~xiaoy/FSDetView/
MIT License
210 stars 33 forks source link

How to construct 200 shots for coco training in phase I? #22

Closed Ze-Yang closed 3 years ago

Ze-Yang commented 3 years ago

Hi,

I am wondering how to construct 200 shots for each class as meta dataset for coco. As the code shows, you only select meta dataset from train2014 only rather than train2014+valminusminival2014. I find that in this way, you are only able to get 44 hair drier and 37 toaster. Even when extend to train2014+valminusminival2014, there are only 198 hair drier available. Could you advise? I am not sure whether I have missed something. Any help is appreciated. Thanks.

Ze-Yang commented 3 years ago

I find that this line with % symbol addresses this problem by recursive sampling, thanks.