0

I have no resource to train the training datasets for Show and Tell: A Neural Image Caption Generator, so where to get the pre-trained models?

JustinGong
  • 399
  • 1
  • 4
  • 17

1 Answers1

1

I haven't tried this, but it seems what you want: https://github.com/KranthiGV/Pretrained-Show-and-Tell-model

I have followed the great Readme at https://github.com/tensorflow/models/tree/master/research/im2txt and am able to build such a model in a few hours on my computer with GTX 1070, the details of which is described in my blog https://jeffxtang.github.io/deep/learning,/hardware,/gpu,/performance/2017/02/14/deep-learning-machine.html in case you're interested in the future.

Ajay Bisht
  • 585
  • 6
  • 8
Jeff Tang
  • 1,804
  • 15
  • 16
  • Hello @Jeff Tang, when you say _able to build in a few hours_ do you refer to building the model with some pre trained weights or training your own weights from scratch? if its the second case, how long is _a few hours_ ? im2txt does says one to two weeks but does not specify its hardware (just GPU, but didn't say nothing about disk, processor and other stuff) Thank you – Rodrigo Laguna Apr 26 '19 at 20:48
  • Sorry @RodrigoLaguna I just noticed your question. I meant running the script under the im2txt link's Initial Training section took me about 3 hours (if I recall correctly). I didn't run or complete the script under the Fine Tune the Inception v3 Model section while the doc says may take several weeks. – Jeff Tang Aug 02 '19 at 17:18