search

tensorflow / models

Models and examples built with TensorFlow
Other
77.16k stars 45.75k forks source link

【Object Detection API】How to evaluate the validation set on multi-GPUs? #1977

Closed wangshy31 closed 7 years ago

wangshy31 commented 7 years ago

【Object Detection API】In terms of training, it is supported for running on multi-GPUs by setting num_clones, but I wonder if it is supported for evaluating the validation set on multi-GPUs. I cannot find related options like "num_clones" in eval.py. Thank you for providing this API!!

drpngx commented 7 years ago

@derekjchow any clue?

derekjchow commented 7 years ago

This currently isn't support by the Tensorflow Object Detection API

wangshy31 commented 7 years ago

Thank you so much! @drpngx @derekjchow

ghost commented 7 years ago

models/object_detection/config/faster_rcnn_resnet101_pets.config batch_size: 2 /models/object_detection/train.py flags.DEFINE_integer(‘num_clones’, 2, ‘Number of clones to deploy per worker.’)

wangshy31 commented 7 years ago

@ideaplat batch_size: 2 and num_clones=2 works for training but not validating period