This project uses CNNs and LSTMs to automatically generate captions from images.
The Microsoft Common Objects in Context (MS COCO) dataset is a large-scale dataset for scene understanding. The dataset is commonly used to train and benchmark object detection, segmentation, and captioning algorithms.
You can read more about the dataset on the website or in the research paper.
To obtain and explore the dataset, you can use the COCO API.
The core architecture is an encoder-decoder, where the encoder is a pretrained ResNet CNN on ImageNet, and the decoder is a basic LSTM.
- Clone this repo: https://github.com/cocodataset/cocoapi
git clone https://github.com/cocodataset/cocoapi.git
- Setup the coco API (also described in the readme here)
cd cocoapi/PythonAPI
make
cd ..
- Download some specific data from here: http://cocodataset.org/#download (described below)
-
Under Annotations, download:
- 2014 Train/Val annotations [241MB] (extract captions_train2014.json and captions_val2014.json, and place at locations cocoapi/annotations/captions_train2014.json and cocoapi/annotations/captions_val2014.json, respectively)
- 2014 Testing Image info [1MB] (extract image_info_test2014.json and place at location cocoapi/annotations/image_info_test2014.json)
-
Under Images, download:
- 2014 Train images [83K/13GB] (extract the train2014 folder and place at location cocoapi/images/train2014/)
- 2014 Val images [41K/6GB] (extract the val2014 folder and place at location cocoapi/images/val2014/)
- 2014 Test images [41K/6GB] (extract the test2014 folder and place at location cocoapi/images/test2014/)
- The project is structured as a series of Jupyter notebooks that are designed to be completed in sequential order (
0_Dataset.ipynb, 1_Preliminaries.ipynb, 2_Training.ipynb, 3_Inference.ipynb).
This is a project from Udacity's computer vision nanodegree
LICENSE: This project is licensed under the terms of the MIT license.