Open HIHIHAHEI opened 4 years ago
Depend on your plateform If u work on google colab It may be take 2and half hour
On Sat, Feb 15, 2020, 13:53 HIHIHAHEI notifications@github.com wrote:
How long does it take to train the model
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/DeepRNN/image_captioning/issues/70?email_source=notifications&email_token=ANWTNZ2ZC6TLZW5BT73EW2LRC6RAHA5CNFSM4KVWQQO2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4INYEW7Q, or unsubscribe https://github.com/notifications/unsubscribe-auth/ANWTNZ5WP2BNN4UHGHR3HATRC6RAHANCNFSM4KVWQQOQ .
It runs very slowly on my own computer, just like the following. Is there any way
batch: 19%|█▊ | 2094/11290 [55:29<4:13:59, 1.66s/it]
Its required more than 4gb GPU for running
On Sat, Feb 15, 2020, 14:25 HIHIHAHEI notifications@github.com wrote:
It runs very slowly on my own computer, just like the following. Is there any way
batch: 19%|█▊ | 2094/11290 [55:29<4:13:59, 1.66s/it]
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/DeepRNN/image_captioning/issues/70?email_source=notifications&email_token=ANWTNZ5LFIEG5ZP7IZHNWM3RC6UWNA5CNFSM4KVWQQO2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEL3FOAA#issuecomment-586569472, or unsubscribe https://github.com/notifications/unsubscribe-auth/ANWTNZ3M33PJ7C6MVE234TDRC6UWNANCNFSM4KVWQQOQ .
My one is 1050ti, 4GB. I'm a student who wants to realize this function, but the cloud server is too expensive
I know that But try to run in google colab Or college system which have gpu
On Sat, Feb 15, 2020, 15:39 HIHIHAHEI notifications@github.com wrote:
My one is 1050ti, 4GB. I'm a student who wants to realize this function, but the cloud server is too expensive
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/DeepRNN/image_captioning/issues/70?email_source=notifications&email_token=ANWTNZZB5GSHBUNY5QJT5TLRC65MZA5CNFSM4KVWQQO2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEL3GXPI#issuecomment-586574781, or unsubscribe https://github.com/notifications/unsubscribe-auth/ANWTNZZK7KA56BYPAQGDSGLRC65MZANCNFSM4KVWQQOQ .
fine, thank u
I'm running the pre-trained model in Colab using CPU (it doesn't seem faster on GPU) and for a single image it takes about 20 seconds to generate the caption and write the result to CSV. I've cached the loaded pre-trained model outside of the session block and disabled generating the image+caption, so it just needs to ingest a single image (see code below).
My implementation is based on Python 3, which I don't think should make a difference. See here: https://github.com/notebookexplore/show-attend-and-tell
Any ideas on how to speed this up or is there a more efficient way to run the pre-trained model?
model = CaptionGenerator(config)
data_dict = np.load('./pre-trained-model/289999.npy', allow_pickle=True, encoding='latin1').item()
with tf.Session() as sess:
# testing phase
data, vocabulary = prepare_test_data(config)
for v in tqdm(tf.global_variables()):
if v.name in data_dict.keys():
sess.run(v.assign(data_dict[v.name]))
model.test(sess, data, vocabulary)
I think, When you are doing this things At every image it will be train the testing phase
On Tue, Mar 3, 2020, 02:14 NotebookExplore notifications@github.com wrote:
I'm running the pre-trained model in Colab using CPU (it doesn't seem faster on GPU) and for a single image it takes about 20 seconds to generate the caption and write the result to CSV. I've cached the loaded pre-trained model outside of the session block and disabled generating the image+caption, so it just needs to ingest a single image (see code below).
Any ideas on how to speed this up or is there a more efficient way to run the pre-trained model?
model = CaptionGenerator(config) data_dict = np.load('./pre-trained-model/289999.npy', allow_pickle=True, encoding='latin1').item() with tf.Session() as sess:
testing phase
data, vocabulary = prepare_test_data(config) for v in tqdm(tf.global_variables()): if v.name in data_dict.keys(): sess.run(v.assign(data_dict[v.name])) model.test(sess, data, vocabulary)
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/DeepRNN/image_captioning/issues/70?email_source=notifications&email_token=ANWTNZ6DUOFMCXLF6XS5K5TRFQLCLA5CNFSM4KVWQQO2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOENQ43VA#issuecomment-593612244, or unsubscribe https://github.com/notifications/unsubscribe-auth/ANWTNZ76VUULEH3HDUPM5OTRFQLCLANCNFSM4KVWQQOQ .
I am using a laboratory server with 4 2080tis and one batch is finishing in 140 minutes. I have to hand in a report about my realizing this function recently so is there anybody who can give some advice about some substitution here for an acceptable result? thx if anybody can give me some advice.
How long does it take to train the model