Facenet: how to use gpu in 'src/compare.py'?

[IvyGongoogle]

hello, I want to use gpu for src/compare.py, and I modify the code as follow:

with tf.Graph().as_default(),tf.device('/gpu:0'):

but it shows error:

compare.py              :56  'main'  in facenet.load_model(model_dir, meta_file, ckpt_file)
facenet.py                    :369  'load_model'  in saver.restore(tf.get_default_session(), os.path.join(model_dir_exp, ckpt_file))
/home/xxx/local/anaconda2/lib/python2.7/site-packages/tensorflow/python/training/saver.py:1428  'restore'  in {self.saver_def.filename_tensor_name: save_path})
/home/xxx/local/anaconda2/lib/python2.7/site-packages/tensorflow/python/client/session.py:767  'run'  in run_metadata_ptr)
/home/xxx/local/anaconda2/lib/python2.7/site-packages/tensorflow/python/client/session.py:965  '_run'  in feed_dict_string, options, run_metadata)
/home/xxx/local/anaconda2/lib/python2.7/site-packages/tensorflow/python/client/session.py:1015  '_do_run'  in target_list, options, run_metadata)
/home/xxx/local/anaconda2/lib/python2.7/site-packages/tensorflow/python/client/session.py:1035  '_do_call'  in raise type(e)(node_def, op, message)

how should I use the gpu in this py to compare two face images?

thank you very much!

[davidsandberg]

Hi @IvyGongoogle,
I usually use the CUDA_VISIBLE_DEVICES environment variable. Have you tried that?

[IvyGongoogle]

@davidsandberg hello, sorry to reply late. Thank you very much!