- Tensorflow implementation of "Show and Tell" http://arxiv.org/abs/1411.4555
- Borrowed some code and ideas from Andrej Karpathy's NeuralTalk.
- You need flickr30k data (images and annotations)
- make_flickr_dataset.py : Extracting feats of flickr30k images, and save them in './data/feats.npy'
- model.py : TensorFlow Version
- Flickr30k Dataset Download
- Extract VGG Featues of Flicker30k images (make_flickr_dataset.py)
- Train: run train() in model.py
- Test: run test() or test_tf() in model.py
- parameters: VGG FC7 feature of test image, trained model path
- Once you download Tensorflow VGG Net (one of the links below), you don't need Caffe when testing.
- Extraced FC7 data: download
- This is used in train() function in model.py. You can skip feature extraction part by using this.
- Pretrained model download
- This is used in test() and test_tf() in model.py. If you do not have time for training, or if you just want to check out captioning, download and test the model.
- Tensorflow VGG net download
- This file is used in test_tf() in model.py
- Along with the files above, you might want to download flickr30k annotation data from link
- BSD license