Human object learning benchmarks
This repository contains code for comparing models of human object learning against measurements of human behavior (n=371k trials).
We developed this repository using Python 3.9. Follow the steps below to use it:
- Begin by cloning the
hobjrepository to your local machine.
- Make and activate a new conda environment.
- Install the
hobjpackage itself. To do so,
hobjdirectory (the top-level one), then run:
pip install -e .
- Then, install the following dependencies:
conda install -c conda-forge xarray dask netCDF4 bottleneck
- Install PyTorch. If you are using a computer without a GPU, you can run the command:
conda install pytorch torchvision torchaudio -c pytorch
Download all images (recommended)
The code in this repository works without this step. But to save time, it is recommended to download the images in a batch. They are located here.
Once it is downloaded, unzip it (it should turn into an
images folder). Move that
images folder to
To see how to view the raw behavioral data and/or score an example learning model, check out the examples in