Pythia

Pythia is a modular framework for Visual Question Answering research, which formed the basis for the winning entry to the VQA Challenge 2018 from Facebook AI Research (FAIR)’s A-STAR team. Please check our paper for more details.

(A-STAR: Agents that See, Talk, Act, and Reason.)

Table of Contents

0. Motivation
1. Citing pythia
2. Installing pythia environment
3. Quick start
4. Preprocess dataset
5. Test with pretrained models
6. Ensemble models
7. Customize config
8. Docker demo
9. AWS s3 dataset summary
10. Acknowledgements
11. References

Motivation

The motivation for Pythia comes from the following observation – a majority of today’s Visual Question Answering (VQA) models fit a particular design paradigm, with modules for question encoding, image feature extraction, fusion of the two (typically with attention), and classification over the space of answers. The long-term goal of Pythia is to serve as a platform for easy and modular research & development in VQA and related directions like visual dialog.

Why the name Pythia?

The name Pythia is an homage to the Oracle of Apollo at Delphi, who answered questions in Ancient Greece. See here for more details.

Citing pythia

If you use Pythia in your research or wish to refer to the baseline results included, please use the following BibTeX entry.

@misc{pythia2018,
  author =       {Yu Jiang and Vivek Natarajan and Xinlei Chen and Marcus Rohrbach and Dhruv Batra and Devi Parikh},
  title =        {Pythia v0.1: The Winning Entry to the VQA Challenge 2018},
  howpublished = {\url{https://github.com/facebookresearch/pythia}},
  year =         {2018}
}

Installing pythia environment

1. Install Anaconda (Anaconda recommended: https://www.continuum.io/downloads).
2. Install cudnn v7.0 and cuda.9.0
3. Create environment for pythia

conda create --name vqa python=3.6

source activate vqa
pip install demjson pyyaml

pip install http://download.pytorch.org/whl/cu90/torch-0.3.1-cp36-cp36m-linux_x86_64.whl

pip install torchvision
pip install tensorboardX

Quick start

We provide preprocessed data files to directly start training and evaluating. Instead of using the original train2014 and val2014 splits, we split val2014 into val2train2014 and minival2014, and use train2014 + val2train2014 for training and minival2014 for validation.

Download data. This step may take some time. Check the sizes of files at the end of readme.

git clone git@github.com:facebookresearch/pythia.git
cd Pythia

mkdir data
cd data
wget https://s3-us-west-1.amazonaws.com/pythia-vqa/data/vqa2.0_glove.6B.300d.txt.npy
wget https://s3-us-west-1.amazonaws.com/pythia-vqa/data/vocabulary_vqa.txt
wget https://s3-us-west-1.amazonaws.com/pythia-vqa/data/answers_vqa.txt
wget https://s3-us-west-1.amazonaws.com/pythia-vqa/data/imdb.tar.gz
wget https://s3-us-west-1.amazonaws.com/pythia-vqa/data/rcnn_10_100.tar.gz
wget https://s3-us-west-1.amazonaws.com/pythia-vqa/data/detectron.tar.gz
wget https://s3-us-west-1.amazonaws.com/pythia-vqa/data/large_vocabulary_vqa.txt
wget https://s3-us-west-1.amazonaws.com/pythia-vqa/data/large_vqa2.0_glove.6B.300d.txt.npy
gunzip imdb.tar.gz 
tar -xf imdb.tar

gunzip rcnn_10_100.tar.gz 
tar -xf rcnn_10_100.tar
rm -f rcnn_10_100.tar

gunzip detectron.tar.gz
tar -xf detectron.tar
rm -f detectron.tar

Optional command-line arguments for train.py

python train.py -h

usage: train.py [-h] [--config CONFIG] [--out_dir OUT_DIR] [--seed SEED]
                [--config_overwrite CONFIG_OVERWRITE] [--force_restart]

optional arguments:
  -h, --help            show this help message and exit
  --config CONFIG       config yaml file
  --out_dir OUT_DIR     output directory, default is current directory
  --seed SEED           random seed, default 1234, set seed to -1 if need a
                        random seed between 1 and 100000
  --config_overwrite CONFIG_OVERWRITE
                        a json string to update yaml config file
  --force_restart       flag to force clean previous result and restart
                        training

Run model without finetuning

cd ../
python train.py

If there is a out of memory error, try:

python train.py --config_overwrite '{data:{image_fast_reader:false}}'

Run model with features from detectron with finetuning

python train.py --config config/keep/detectron.yaml

Check result for the default run

cd results/default/1234

The results folder contains the following info

results
|_ default
|  |_ 1234 (default seed)
|  |  |_config.yaml
|  |  |_best_model.pth
|  |  |_best_model_predict_test.pkl 
|  |  |_best_model_predict_test.json (json file for predicted results on test dataset)
|  |  |_model_00001000.pth (mpdel snapshot at iter 1000)
|  |  |_result_on_val.txt
|  |  |_ ...
|  |_(other_cofig_setting)
|  |  |_...
|_ (other_config_file)
|

The log files for tensorbord are stored under boards/

Preprocess dataset

If you want to start from the original VQA dataset and preprocess data by yourself, use the following instructions in data_preprocess.md. This part is not necessary if you download all data from quick start.

Test with pretrained models

Note: all of these models below are trained with validation set included

Description	performance (test-dev)	Link
detectron_100_resnet_most_data	70.01	https://s3-us-west-1.amazonaws.com/pythia-vqa/pretrained_models/detectron_100_resnet_most_data.tar.gz
baseline	68.05	https://s3-us-west-1.amazonaws.com/pythia-vqa/pretrained_models/baseline.tar.gz
baseline +VG +VisDal +mirror	68.98	https://s3-us-west-1.amazonaws.com/pythia-vqa/pretrained_models/most_data.tar.gz
detectron_finetune	68.49	https://s3-us-west-1.amazonaws.com/pythia-vqa/pretrained_models/detectron.tar.gz
detectron_finetune+VG +VisDal +mirror	69.24	https://s3-us-west-1.amazonaws.com/pythia-vqa/pretrained_models/detectron_most_data.tar.gz

Best Pretrained Model

The best pretrained model can be downloaded as follows:

mkdir pretrained_models/
cd pretrained_models
wget https://s3-us-west-1.amazonaws.com/pythia-vqa/pretrained_models/detectron_100_resnet_most_data.tar.gz
gunzip detectron_100_resnet_most_data.tar.gz 
tar -xf detectron_100_resnet_most_data.tar
rm -f detectron_100_resnet_most_data.tar

Get ResNet152 features and Detectron features with fixed 100 bounding boxes

cd data
wget https://s3-us-west-1.amazonaws.com/pythia-vqa/data/detectron_fix_100.tar.gz
gunzip detectron_fix_100.tar.gz
tar -xf detectron_fix_100.tar
rm -f detectron_fix_100.tar

wget https://s3-us-west-1.amazonaws.com/pythia-vqa/data/resnet152.tar.gz
gunzip resnet152.tar.gz
tar -xf resnet152.tar
rm -f resnet152.tar

Test the best model on the VQA test2015 dataset

python run_test.py --config pretrained_models/detectron_100_resnet_most_data/1234/config.yaml \
--model_path pretrained_models/detectron_100_resnet_most_data/1234/best_model.pth \
--out_prefix test_best_model

The results will be saved as a json file test_best_model.json, and this file can be uploaded to the evaluation server on EvalAI (https://evalai.cloudcv.org/web/challenges/challenge-page/80/submission).

Ensemble models

Download all the models above

python ensemble.py --res_dirs pretrained_models/ --out ensemble_5.json

Results will be saved in ensemble_5.json. This ensemble can get accuracy 71.65 on test-dev.

Ensemble 30 models

To run an ensemble of 30 pretrained models, download the models and image features as follows. This gets an accuracy of 72.18 on test-dev.

wget https://s3-us-west-1.amazonaws.com/pythia-vqa/ensembled.tar.gz

Customize config

To change models or adjust hyper-parameters, see config_help.md

Docker demo

To quickly tryout a model interactively with nvidia-docker

git clone https://github.com/facebookresearch/pythia.git
nvidia-docker build pythia -t pythia:latest
nvidia-docker run -ti --net=host pythia:latest

This will open a jupyter notebook with a demo model to which you can ask questions interactively.

AWS s3 dataset summary

Here, we listed the size of some large files in our AWS S3 bucket.

Description	size
data/rcnn_10_100.tar.gz	71.0GB
data/detectron.tar.gz	106.2 GB
data/detectron_fix_100.tar.gz	162.6GB
data/resnet152.tar.gz	399.6GB
ensembled.tar.gz	462.1GB

Acknowledgements

We would like to thank Peter Anderson, Abhishek Das, Stefan Lee, Jiasen Lu, Jianwei Yang, Licheng Yu, Luowei Zhou for helpful discussions, Peter Anderson for providing training data for the Visual Genome detector, Deshraj Yadav for responses on EvalAI related questions, Stefan Lee for suggesting the name Pythia, Meet Shah for building the docker demo for Pythia and Abhishek Das, Abhishek Kadian for feedback on our codebase

References

• P. Anderson, X. He, C. Buehler, D. Teney, M. Johnson, S. Gould, and L. Zhang. Bottom-up and top-down attenttion for image captioning and visual question answering. In CVPR, 2018.
• S. Antol, A. Agrawal, J. Lu, M. Mitchell, D. Batra,C. Lawrence Zitnick, and D. Parikh. VQA: Visual question answering. In ICCV, 2015
• A. Das, S. Kottur, K. Gupta, A. Singh, D. Yadav, J. M.Moura, D. Parikh, and D. Batra. Visual Dialog. In CVPR,2017
• Y. Goyal, T. Khot, D. Summers-Stay, D. Batra, and D. Parikh. Making the V in VQA matter: Elevating the role of image understanding in Visual Question Answering. In CVPR, 2017.
• Hu R, Andreas J, Rohrbach M, Darrell T, Saenko K. Learning to reason: End-to-end module networks for visual question answering. In ICCV, 2017.
• D. Teney, P. Anderson, X. He, and A. van den Hengel. Tips and tricks for visual question answering: Learnings from the 2017 challenge. CoRR, abs/1708.02711, 2017.
• Z. Yu, J. Yu, C. Xiang, J. Fan, and D. Tao. Beyond bilinear: Generalized multimodal factorized high-order pooling for visual question answering. TNNLS, 2018.