Reproduce result

[WangYueFt]

Hi Kelvin,

After I trained model 2000 epochs, I tried to use generate_caps.py to get the captions. However, the result was meaningless. Is there anything I need to take care of regarding the generate_caps.py?

Thanks

[intuinno]

Hello Wang,

I want to help you. I am at least getting meaningful captions. However your question is still too vague and there can be many things that can be broken. I suggest you look at my fork https://github.com/intuinno/arctic-captions , which has all the required code but with poor documentation.

Thanks.

[WangYueFt]

@intuinno Thank you very much! I found the way I generated training data was wrong. I am looking into your code. Did you reproduce the flickr8k data? And how was the result?

Thanks, Yue

[intuinno]

@WangYueFt I did not reproduce flickr8k. Please use code for coco data set because they are the only one most up to date.

Thanks.

[2g-XzenG]

Hello @intuinno , I try your code, and I am a little confuse about path in your prepare_coco.py, vgg_deploy_path = '/home/intuinno/codegit/caffe/models/vgg_ilsvrc_19/VGG_ILSVRC_19_layers_deploy.prototxt' vgg_model_path = '/home/intuinno/codegit/caffe/models/vgg_ilsvrc_19/VGG_ILSVRC_19_layers.caffemodel' can you explain a little bit about these two? Thank you very much!

[AAmmy]

@wahahayawahaha You can google those names and download them.