Question :: Python 3.7 with tensorflow?

[zbeedatm]

Hi,
I tried installing tensorflow on top of python 3.7 but didn't succeed and I understood that 3.7 still not supporting tensorflow and there is a need to downgrade to 3.6.

I had to make some changes on the code from this repo to make it compile and work with python 3.6

If there is a possibility to run your code as is (python 3.7 I guess) with tensorflow please let me know what am I missing.

Thanks.

[ThatNight]

I can run it by python3.6 and tensorflow1.2 without modification

[zbeedatm]

Weird, I have:
Python 3.6.3 (v3.6.3:2c5fed8, Oct 3 2017, 18:11:49) [MSC v.1900 64 bit (AMD64)] on win32

But some code was failing in run-time! like:

base-model:
fp = open(config.eval_result_file, 'w') #w instead of wb: TypeError: a bytes-like object is required, not 'str'

#TODO
data_dict = np.load(data_path, allow_pickle=True, encoding='latin1').item()

in many places had to switch the for loop suntax:

#for param_name, data in data_dict[op_name].iteritems():
for param_name, data in data_dict[op_name].items():

issues like that...

tensorflow:
Name: tensorflow
Version: 1.14.0

Anyways, when I took the trained model and ran the code after fixing all issues I'm not getting good results!
You wrote that you trained it with default configurations, then I guess it was trained with VGG16.
Is there a chance that you have a better trained model that you can share?

What configurations can be changed to get better results in general?

It will help me a lot.

Thanks in advance.

[ThatNight]

Hello, after I run the program, I also found that the results are not satisfactory. I'm also trying to figure out how to make the model work better. You can try to change the network structure of Lstm.you can see the Geogle open source project im2txt, the effect is good. If you have a better way to get better results, I hope you can tell me. Thank you in advance.

…

------------------ 原始邮件 ------------------ 发件人: "zbeedatm"<[email protected]>; 发送时间: 2019年8月1日(星期四) 下午2:37 收件人: "DeepRNN/image_captioning"<[email protected]>; 抄送: "冰炫绿忆"<[email protected]>; "Comment"<[email protected]>; 主题: Re: [DeepRNN/image_captioning] Question :: Python 3.7 withtensorflow? (#61) Weird, I have: Python 3.6.3 (v3.6.3:2c5fed8, Oct 3 2017, 18:11:49) [MSC v.1900 64 bit (AMD64)] on win32 But some code was failing in run-time! like: base-model: fp = open(config.eval_result_file, 'w') #w instead of wb: TypeError: a bytes-like object is required, not 'str' #TODO data_dict = np.load(data_path, allow_pickle=True, encoding='latin1').item() in many places had to switch the for loop suntax: #for param_name, data in data_dict[op_name].iteritems(): for param_name, data in data_dict[op_name].items(): issues like that... tensorflow: Name: tensorflow Version: 1.14.0 Anyways, when I took the trained model and ran the code after fixing all issues I'm not getting good results! You wrote that you trained it with default configurations, then I guess it was trained with VGG16. Is there a chance that you have a better trained model that you can share? What configurations can be changed to get better results in general? It will help me a lot. Thanks in advance. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

[zbeedatm]

The problem that I don' have a strong enough machine :S
I have a laptop with 1 GPU & windows on it, and when running the training (with CPU actually, didn't try with GPU yet) it was crashing after one day. I guess with GPU will work better... does your code support running under GPU? or it's a matter of installing the GPU tensorflow?

I'm not familiar..

[ThatNight]

I'm a model trained on the Tensorflow GPU. But the effect is not very good. I have a problem now. Do you know the Beam Search method in BaseModel? I don't understand what the code means. If you know every line of Beam_Search method code, can you write me a comment on each line of code? Thank you

…

------------------ 原始邮件 ------------------ 发件人: "zbeedatm"<[email protected]>; 发送时间: 2019年8月1日(星期四) 下午5:49 收件人: "DeepRNN/image_captioning"<[email protected]>; 抄送: "冰炫绿忆"<[email protected]>; "Comment"<[email protected]>; 主题: Re: [DeepRNN/image_captioning] Question :: Python 3.7 withtensorflow? (#61) The problem that I don' have a strong enough machine :S I have a laptop with 1 GPU & windows on it, and when running the training (with CPU actually, didn't try with GPU yet) it was crashing after one day. I guess with GPU will work better... does your code support running under GPU? or it's a matter of installing the GPU tensorflow? I'm not familiar.. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

[maxy0524]

https://github.com/79837383/image_captioning
I modified the code, support python3.6, python3.7,TensorFlow1.8 TensorFlow1.12 TensorFlow1.13 TensorFlow1.14, numpy 1.12 or newer, and Baidu disk download link if you like.