shawnking98 / image-caption-with-attention

Image-Caption-with-Attention

Pytorch implementation of the paper Show, Attend and Tell: Neural Image Caption Generation with Visual Attention.

UMich Fall 2021 EECS 545 final project.

Running guide:

• Download dataset from COCO2014 or Vizwiz
• Download the python api of COCO or Vizwiz
• Run prepro.py to build the vocabulary and resize the pictures
• Run main.py to start training
• Run validate.py along with passed-in checkpoint & vocabulary path to do the test. This script can predict a caption given a image, as well as give bleu scores on given validation set.

Our network parameters and vocabulary file for both datasets have been uploaded to Google drive, trained for 3 days on COCO (~30 epoches) and Vizwiz (~100 epoches) using Greatlakes servers.