Skip to content

A Diagnostic Dataset for Compositional Language and Elementary Visual Reasoning

License

Notifications You must be signed in to change notification settings

zerolatnc/clevr-dataset-gen

 
 

Repository files navigation

CLEVR Dataset Generation

This is the code used to generate the CLEVR dataset as described in the paper:

CLEVR: A Diagnostic Dataset for Compositional Language and Elementary Visual Reasoning
Justin Johnson, Bharath Hariharan, Laurens van der Maaten, Fei-Fei Li, Larry Zitnick, Ross Girshick
Presented at CVPR 2017

Code and pretrained models for the baselines used in the paper can be found here.

You can use this code to render synthetic images and compositional questions for those images, like this:

Q: How many small spheres are there?
A: 2

Q: What number of cubes are small things or red metal objects?
A: 2

Q: Does the metal sphere have the same color as the metal cylinder?
A: Yes

Q: Are there more small cylinders than metal things?
A: No

Q: There is a cylinder that is on the right side of the large yellow object behind the blue ball; is there a shiny cube in front of it?
A: Yes

If you find this code useful in your research then please cite

@inproceedings{johnson2017clevr,
  title={CLEVR: A Diagnostic Dataset for Compositional Language and Elementary Visual Reasoning},
  author={Johnson, Justin and Hariharan, Bharath and van der Maaten, Laurens
          and Fei-Fei, Li and Zitnick, C Lawrence and Girshick, Ross},
  booktitle={CVPR},
  year={2017}
}

All code was developed and tested on OSX and Ubuntu 16.04.

Step 1: Generating Images

First we render synthetic images using Blender, outputting both rendered images as well as a JSON file containing ground-truth scene information for each image.

Blender ships with its own installation of Python which is used to execute scripts that interact with Blender; you'll need to add the image_generation directory to Python path of Blender's bundled Python. The easiest way to do this is by adding a .pth file to the site-packages directory of Blender's Python, like this:

echo $PWD/image_generation >> $BLENDER/$VERSION/python/lib/python3.5/site-packages/clevr.pth

where $BLENDER is the directory where Blender is installed and $VERSION is your Blender version; for example on OSX you might run:

echo $PWD/image_generation >> /Applications/blender/blender.app/Contents/Resources/2.78/python/lib/python3.5/site-packages/clevr.pth

You can then render some images like this:

cd image_generation
blender --background --python render_images.py -- --num_images 10

On OSX the blender binary is located inside the blender.app directory; for convenience you may want to add the following alias to your ~/.bash_profile file:

alias blender='/Applications/blender/blender.app/Contents/MacOS/blender'

If you have an NVIDIA GPU with CUDA installed then you can use the GPU to accelerate rendering like this:

blender --background --python render_images.py -- --num_images 10 --use_gpu 1

After this command terminates you should have ten freshly rendered images stored in output/images like these:


The file output/CLEVR_scenes.json will contain ground-truth scene information for all newly rendered images.

You can find more details about image rendering here.

Step 2: Generating Questions

Next we generate questions, functional programs, and answers for the rendered images generated in the previous step. This step takes as input the single JSON file containing all ground-truth scene information, and outputs a JSON file containing questions, answers, and functional programs for the questions in a single JSON file.

You can generate questions like this:

cd question_generation
python generate_questions.py

The file output/CLEVR_questions.json will then contain questions for the generated images.

You can find more details about question generation here.

About

A Diagnostic Dataset for Compositional Language and Elementary Visual Reasoning

Resources

License

Code of conduct

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%