adrianmarino/image-captioning

Image Captioning

Requeriments

• anaconda/anaconda-natigator
• 3D video card (i.e. GeForce GTX 1060 or higher)

Used datasets & world embeddings

• Datasets
  • Flickr 8k
  • Flickr 30k
• Word embedings
  • Glove word embedding
    • glove.6B
    • glove.840B.300d
  • Elmo embedding

Setup environment

Step 1: Create project environment.

$ conda env create --file environment.yml

Step 2: Activate environment.

$ conda activate image-captioning

Prepare dataset

Step 1: Download & prepare word embeddings to use:

$ ./setup_word_embedings

Step 2: Download Flickr8K dataset .torrent from this.

Step 3: Once downloaded the Flickr directory, make dataset directory under project path and copy downloaded directory to this.

$ mkdir -p dataset
$ cp -rf ~/Downloads/Flickr8k ~/project/dataset/flickr8k

Step 4: Login an download Flickr30k dataset from kaggle.

Step 5: Copy downloaded file to project/dataset path and unzip this.

$ cp -rf ~/Downloads/flickr30k_images.zip ~/project/dataset/
$ cd ~/project/dataset
$ unzip flickr30k_images.zip

Step 6: Finally change dataset directory structure like this.

Train/Test model

You can train, test and adjust model from Image captioning notebook.

Word embeddings

Word embedding use examples: word_embedding_examples.