Awesome
show-adapt-and-tell
This is the official code for the paper
Show, Adapt and Tell: Adversarial Training of Cross-domain Image Captioner <br> Tseng-Hung Chen, Yuan-Hong Liao, Ching-Yao Chuang, Wan-Ting Hsu, Jianlong Fu, Min Sun <br> To appear in ICCV 2017
<div align='center'> <img src='images/teaser.png' height="405px"> </div>In this repository we provide:
- The cross-domain captioning models used in the paper
- Script for preprocessing MSCOCO data
- Script for preprocessing CUB-200-2011 captions
- Code for training the cross-domain captioning models
If you find this code useful for your research, please cite
@article{chen2017show,
title={Show, Adapt and Tell: Adversarial Training of Cross-domain Image Captioner},
author={Chen, Tseng-Hung and Liao, Yuan-Hong and Chuang, Ching-Yao and Hsu, Wan-Ting and Fu, Jianlong and Sun, Min},
journal={arXiv preprint arXiv:1705.00930},
year={2017}
}
Requirements
- Python 2.7
- Tensoflow 0.12.1
- Caffe
- OpenCV 2.4.9
P.S. Please clone the repository with the --recursive
flag:
# Make sure to clone with --recursive
git clone --recursive https://github.com/tsenghungchen/show-adapt-and-tell.git
Data Preprocessing
MSCOCO Captioning dataset
Feature Extraction
- Download the pretrained ResNet-101 model and place it under
data-prepro/MSCOCO_preprocess/resnet_model/
. - Please modify the caffe path in
data-prepro/MSCOCO_preprocess/extract_resnet_coco.py
. - Go to
data-prepro/MSCOCO_preprocess
and run the following script:./download_mscoco.sh
for downloading images and extracting features.
Captions Tokenization
- Clone the NeuralTalk2 repository and head over to the coco/ folder and run the IPython notebook to generate a json file for Karpathy split:
coco_raw.json
. - Run the following script:
./prepro_mscoco_caption.sh
for downloading and tokenizing captions. - Run
python prepro_coco_annotation.py
to generate annotation json file for testing.
CUB-200-2011 with Descriptions
Feature Extraction
- Run the script
./download_cub.sh
to download the images in CUB-200-2011. - Please modify the input/output path in
data-prepro/MSCOCO_preprocess/extract_resnet_coco.py
to extract and pack features in CUB-200-2011.
Captions Tokenization
- Download the description data.
- Run
python get_split.py
to generate dataset split following the ECCV16 paper "Generating Visual Explanations". - Run
python prepro_cub_annotation.py
to generate annotation json file for testing. - Run
python CUB_preprocess_token.py
for tokenization.
Models from the paper
Pretrained Models
Download all pretrained and adaption models:
- MSCOCO pretrained model
- CUB-200-2011 adaptation model
- TGIF adaptation model
- Flickr30k adaptation model
Example Results
Here are some example results where the captions are generated from these models:
<table border=1> <tr> <td> <img src='images/im6795.jpg' height="185px"> </br> MSCOCO: A large air plane on a run way. </br> CUB-200-2011: A large white and black airplane with a large beak. </br> TGIF: A plane is flying over a field. </br> Flickr30k: A large airplane is sitting on a runway. </td> <td> <img src='images/im11063.jpg' height="185px"> </br> MSCOCO: A traffic light is seen in front of a large building. </br> CUB-200-2011: A yellow traffic light with a yellow light. </br> TGIF: A traffic light is hanging on a pole. </br> Flickr30k: A street sign is lit up in the dark </td> </tr> </table> <table border=1> <tr> <td> <img src='images/im22197.jpg' height="185px"> </br> MSCOCO: A black dog sitting on the ground next to a window. </br> CUB-200-2011: A black and white dog with a black head. </br> TGIF: A dog is looking at something in the mirror. </br> Flickr30k: A black dog is looking out of the window. </td> <td> <img src='images/im270.jpg' height="185px"> </br> MSCOCO: A man riding a skateboard up the side of a ramp. </br> CUB-200-2011: A man riding a skateboard on a white ramp. </br> TGIF: A man is doing a trick on a skateboard. </br> Flickr30k: A man in a blue shirt is doing a trick on a skateboard. </td> </tr> </table>Training
The training codes are under the show-adapt-tell/
folder.
Simply run python main.py
for two steps of training:
Training the source model with paired image-caption data
Please set the Boolean value of "G_is_pretrain"
to True in main.py
to start pretraining the generator.
Training the cross-domain captioner with unpaired data
After pretraining, set "G_is_pretrain"
to False to start training the cross-domain model.
License
Free for personal or research use; for commercial use please contact me.