iWISDM

iWISDM, short for instructed-Virtual VISual Decision Making, is a virtual environment capable of generating a limitless array of vision-language tasks with varying complexity. iWISDM encompasses a broad spectrum of tasks that engage executive functions such as inhibition of action, working memory, attentional set, task switching, and schema generalization. It is also a scalable and extensible framework which allows users to easily define their own task space and stimuli dataset. iWISDM builds on the compositional nature of human behavior, and the fact that complex tasks are often constructed by combining smaller task units together in time.

Below is an example of the generated tasks:

iWISDM inherits several classes from COG (https://github.com/google/cog) to build task graphs. For convenience, we have pre-implemented several commonly used cognitive tasks in task_bank.py.

For usage instructions, please refer to Usage

Additionally, for convenience, we have pre-generated four benchmarks of increased complexity level for evaluation of large multi-modal models.

These datasets can be generated from /benchmarking or downloaded: iWISDM_benchsets.tar.gz

For further details, please refer to (https://arxiv.org/submit/5678755/view)

Usage

Install Instructions

Graphiz

• Install graphiz on your machine

Poetry

Install Poetry

curl -sSL https://install.python-poetry.org | python3 -

Conda + Poetry

Create conda python environment

conda create --name iwisdm python=3.11

Install packages

poetry install

ShapeNet Subset

A large-scale repository of shapes represented by 3D CAD models of objects (Chang et. al. 2015).

Pre-rendered Dataset Download

shapenet_handpicked.tar.gz

Basic Usage

# imports
from wisdom import make
from wisdom import read_write

# environment initialization
with open('../benchmarking/configs/high_complexity_all.json', 'r') as f:
    config = json.load(f)  # using pre-defined AutoTask configuration
env = make(env_id='ShapeNet')
env.set_env_spec(
    env.init_env_spec(
        auto_gen_config=config,
    )
)

# AutoTask procedural task generation and saving trial
tasks = env.generate_tasks(10)  # generate 10 random task graphs and tasks
_, (_, temporal_task) = tasks[0]
trials = env.generate_trials(tasks=[temporal_task])  # generate a trial
imgs, _, info_dict = trials[0]
read_write.write_trial(imgs, info_dict, f'output/trial_{i}')

See /tutorials for more examples.

Acknowledgements

This repository builds upon the foundational work presented in the COG paper (https://arxiv.org/abs/1803.06092).

Yang, Guangyu Robert, et al. "A dataset and architecture for visual reasoning with a working memory." Proceedings of the European Conference on Computer Vision (ECCV). 2018.

Citation

If you find iWISDM useful in your research, please use the following BibTex:

@inproceedings{lei2024iwisdm,
  title={iWISDM: Assessing instruction following in multimodal models at scale},
  author={Lei, Xiaoxuan and Gomez, Lucas and Bai, Hao Yuan and Bashivan, Pouya},
  booktitle={Conference on Lifelong Learning Agents (CoLLAs 2024)},
  year={2024}
}