Large-scale electrophysiological recordings from V1, V4 and IT in two macaques in response to ~22k images from the THINGS image database.
Paolo Papale, Feng Wang, Matthew W. Self and Pieter R. Roelfsema
Dept. of Vision and Cognition, Netherlands Institute for Neuroscience (KNAW), 1105 BA Amsterdam (NL).
N.B. The THINGS stimuli from Martin Hebart (et al.) are not provided, but you can download them on things-initiative.org
A full description of the dataset is provided in the Neuron paper. A few additional things are needed to be able to work on the data.
Data for each monkey is provided in a folder, containing both the RAW and MUA data. The RAW data is subdived in different days of recordings, and individual blocks/runs of ~20 minutes of lenght. We provide the MATLAB code to extract the MUA out of RAW data, aggregate all the trials across blocks and days, normalize it, filter and chunck it for model training. These scripts can be easily changed to extract LFP, or to aggregate the RAW data, or look into not-completed images.
These scripts can be found in "_code" and are (in sequence):
The MUA data is provided both un-normalized ("THINGS_MUA_trials.mat") and normalized and averaged in time-windows ("THINGS_normMUA.mat")
THINGS_MUA_trials.mat contains:
THINGS_normMUA.mat contains:
The scripts and data rely on logfiles hosted in "_logs". There, you can also find "things_imgs.mat" that is required to associate each stimulus from the THINGS initiative database to the specific trial (see below). things_imgs.mat contains:
N.B. the normalized data is already sorted according to the order of images in "train_imgs" and "test_imgs" from "things_imgs.mat"
In addition to the scripts mentioned, we provide the Matlab APIs from Blackrock Neurotech, the same version used for the paper. Also, we provide a few util functions that are called by the main scripts or can be used to plot some of the results. Finally, we provide the Python code for the MEIs in "lucent-things", based on the lucent viz library GitHub link
Create an account on gin and download the gin client as described here. On your computer, log in using:
gin login
Clone the repository using:
gin get paolo_papale/TVSD
The cloning step can take a long time, due to the large amount of individual files. Please be patient.
Large data files will not be downloaded automatically, they will appear as git-annex links instead. We recommend downloading only the files you need, since the entire dataset is large. To get the contents of a certain file:
gin get-content <filename>
Downloaded large files will be read-only. You might want to unlock the files using:
gin unlock <filename>
To remove the contents of a large file again, use:
gin remove-content <filename>
Detailed description of the gin client can be found at the gin wiki. See the gin usage tutorial for advanced features.
Download the files you want by clicking download in the gin web interface.
If you are interested in modeling/tuning, you can download the normalized MUA for each monkey by clicking here:
monkey N: https://gin.g-node.org/paolo_papale/TVSD/raw/master/monkeyN/THINGS_normMUA.mat
monkey F: https://gin.g-node.org/paolo_papale/TVSD/raw/master/monkeyF/THINGS_normMUA.mat
Cite this work by citing the original publication.
Don't use our institutional emails for questions about the TVSD, instead you can reach us at things [dot] tvsd [at] gmail [dot] com
The data and metadata in this work are licensed under a Creative Commons Attribution 4.0 International License.
Python (v. 2.7) and Matlab (v. 2019b) code in this repository are licensed under the same license, with the following exceptions:
The matlab-based NPMK package provided within this repository is re-distributed under the BSD 3-clause license, in compliance with the original licensing terms.
The python-based lucent library provided within this repository is re-distributed under the Apache License 2.0 license, in compliance with the original licensing terms.
The python-based models under lucent provided within this repository are re-distributed under the MIT License license, in compliance with the original licensing terms.