THINGSvision: A Python Toolbox for Streamlining the Extraction of Activations From Deep Neural Networks

Abstract

Over the past decade, deep neural network (DNN) models have received a lot of attention due to their near-human object classification performance and their excellent prediction of signals recorded from biological visual systems. To better understand the function of these networks and relate them to hypotheses about brain activity and behavior, researchers need to extract the activations to images across different DNN layers. The abundance of different DNN variants, however, can often be unwieldy, and the task of extracting DNN activations from different layers may be non-trivial and error-prone for someone without a strong computational background. Thus, researchers in the fields of cognitive science and computational neuroscience would benefit from a library or package that supports a user in the extraction task. THINGSvision is a new Python module that aims at closing this gap by providing a simple and unified tool for extracting layer activations for a wide range of pretrained and randomly-initialized neural network architectures, even for users with little to no programming experience. We demonstrate the general utility of THINGsvision by relating extracted DNN activations to a number of functional MRI and behavioral datasets using representational similarity analysis, which can be performed as an integral part of the toolbox. Together, THINGSvision enables researchers across diverse fields to extract features in a streamlined manner for their custom image dataset, thereby improving the ease of relating DNNs, brain activity, and behavior, and improving the reproducibility of findings in these research fields.

Perceptual reorganisation from prior knowledge emerges late in childhood

“…Pretrained AlexNet and NASNet-large models were acquired and run on Matlab 2020a's Deep Learning Toolbox (MathWorks, Natick, MA, USA). CorNet models (RT, Z, S) were acquired and run with the Python toolbox THINGSvision (Muttenthaler and Hebart, 2021). Due to poorer greyscale performance, CorNet-RT and CorNet-Z were not included in the results.…”

TorchLens: A Python package for extracting and visualizing hidden activations of PyTorch models

“…In addition to these manual approaches for extracting intermediate activations from DNNs, three open-source packages now exist for facilitating this process. ThingsVision (Muttenthaler & Hebart, 2021) is a Python package with user-friendly functionality for loading pre-existing DNNs, extracting their activations either to common image datasets or to user-defined datasets, and performing various common analyses on the extracted features, such as representational similarity analysis (Kriegeskorte et al, 2008). While it can extract activations from a large and convenient library of pre-existing models, it does not automatically work for arbitrary new models (e.g., user-designed models), and can only extract the outputs of PyTorch modules, not every operation in the model.…”

Computational Models Suggest that Human Memory Judgments Exhibit Interference due to the Use of Distributed Representations

“…However, we think that it would also be interesting to study the relationship between a measure of stimulus similarity (i.e., a physical or semantic measure of similarity between targets and lures) to compare the transfer function between this objective measure of stimulus similarity and the degree of memory interference for the similar lures. Accordingly, we extracted features (i.e., modeled neural responses) from the 3rd convolutional layer of IT (i.e., the penultimate layer) from CORnet-S [60] using the THINGSvision toolbox [64]. Briefly, CORnet-S has been shown to provide a good approximation both to neural responses in IT as well as to behavioral data across several studies [60,64].…”