markdtw / soft-attention-image-captioning Goto Github PK

Thank you for your code which is really helpful for my learning.
In the function that calculates the cross-entropy loss over a candidate sentence, what is the purpose of the variable "mask_ph"? The loss tensor is raised to the power of mask_ph, and afterwards divided by the sum of mask_ph.

Best wishes
James

Download and extract pre-trained Inception V4 and VGG 19 from tf.slim for feature extraction.
Save the ckpt files in cnns/ as inception_v4_imagenet.ckpt and vgg_19_imagenet.ckpt.

Thank you for the great implementation.
Is there a pre-trained model available for the captioner?
Thanks again!

As the title suggest, I am getting a MemoryError when trying to save the features in extract_all.
I have an i7-6402p, gtx 1060gb, and 8gb of ram. The next question might be silly, as I am just a beginner.

I was running utils.py with tensorflow on the gpu. I am assuming the RAM memory is used when saving the features. As such, if I were to run utils.py with tensorflow on the CPU, am I correct to say that I would get the same error when attempting to save the file ?
I should mention that on the GPU I do receive some notification / errors, which do not show up on the cpu.

How could I fix this issue ? Would it be possible for you to share the npy file ?

Thank you for your time and I await your response.