https://github.com/iamshnoo/udacity-cvnd-p2---image-captioning

Uses a CNN Encoder and a RNN Decoder to generate captions for input images.
https://github.com/iamshnoo/udacity-cvnd-p2---image-captioning

Last synced: 8 days ago
JSON representation

Uses a CNN Encoder and a RNN Decoder to generate captions for input images.

• Host: GitHub
• URL: https://github.com/iamshnoo/udacity-cvnd-p2---image-captioning
• Owner: iamshnoo
• License: mit
• Created: 2019-03-21T13:53:35.000Z (over 5 years ago)
• Default Branch: master
• Last Pushed: 2020-10-31T21:29:15.000Z (about 4 years ago)
• Last Synced: 2023-03-07T11:33:48.012Z (over 1 year ago)
• Language: Jupyter Notebook
• Homepage:
• Size: 2.75 MB
• Stars: 2
• Watchers: 1
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# UDACITY-CVND-P2---Image-Captioning

[![Udacity Computer Vision Nanodegree](http://tugan0329.bitbucket.io/imgs/github/cvnd.svg)](https://www.udacity.com/course/computer-vision-nanodegree--nd891)


Uses a CNN Encoder and a RNN Decoder to generate captions for input images.


The Project has been reviewed by Udacity and graded Meets Specifications. 


Here's a sumary of the steps involved.

- Dataset used is the COCO data set by Microsoft.

- Feature vectors for images are generated using a CNN based on the ResNet architecture by Google.

- Word embeddings are generated from captions for training images. NLTK was used for working with processing of captions.

- Implemented an RNN decoder using LSTM cells.

- Trained the network for nearly 3 hrs using GPU to achieve average loss of about 2%.

- Obtained outputs for some test images to understand efficiency of the trained network.


![Alt](https://raw.githubusercontent.com/udacity/CVND---Image-Captioning-Project/master/images/encoder-decoder.png)