search

Anjaney1999 / image-captioning-seqgan

An image captioning model that is inspired by the Show, Attend and Tell paper (https://arxiv.org/abs/1502.03044) and the Sequence Generative Adversarial Network paper (https://arxiv.org/abs/1609.05473)
22 stars 2 forks source link

which subsets from the datasets (flickr8k, flickr30k, coco) put int the right place #1

Open simohamedhdafa opened 4 years ago

simohamedhdafa commented 4 years ago

In the second step to run the program : Place images from flickr8k, flickr30, or coco within the images folder (make sure to place it in the correct subdirectory)

Do I need to place only training sets (from each dataset) in the right place ? or mixed train/val/test sets (from each dataset) in the right place ? is it related to karpathy_splits ? if yes, how ?

Thank you

simohamedhdafa commented 4 years ago

I notice the line 181 in "preprocess.py" file : parser.add_argument('--dataset', type=str, default='flickr8k') I assume that in this line we can specify which dataset to use

simohamedhdafa commented 4 years ago

The mistakes I've made helped me to understand that we have to put the classical "train2014" and "val2014" folders (from the MSCOCO dataset) within the "coco" folder. If this can help anyone !

Anjaney1999 commented 4 years ago

Oh shoot, I am so sorry. I have been a bit busy these past couple of days. Yes, you're right. I have also corrected a few logical errors I found in the code, so please check it out if you have the time:)

Zyg187 commented 5 months ago

Your code is really neat and beautiful. I really like your code style. It would be great if you could provide a more detailed introduction. You are really great.