search

zengarden / light_head_rcnn

Light-Head R-CNN
834 stars 222 forks source link

it seems that you did not use ohem? #71

Open qimw opened 5 years ago

qimw commented 5 years ago

In the config.py, the TRAIN.nr_ohem_sampling = 256 * train_batch_per_gpu. The TRAIN.RPN_BATCHSIZE is also 256. Did you really use ohem in your code ?