How do we recognize objects, make sense of them, and act on them meaningfully? These fundamental questions require collaboration across disciplines - psychology, neuroscience, and AI.
THINGS provides a foundation: 1,854 systematically-sampled object concepts with 26,107 naturalistic images, plus millions of behavioral judgments and neural recordings from multiple species and brain imaging techniques.
Labs worldwide contribute data openly. By working from the same objects, we can finally bridge the gap between brain and behavior. Anyone can join the initiative.
THINGS
A global research initiative built on a shared image database
How do we recognize objects, make sense of them, and act on them meaningfully? These fundamental questions require collaboration across disciplines - psychology, neuroscience, and AI.
THINGS provides a foundation: 1,854 systematically-sampled object concepts with 26,107 naturalistic images, plus millions of behavioral judgments and neural recordings from multiple species and brain imaging techniques.
Labs worldwide contribute data openly. By working from the same objects, we can finally bridge the gap between brain and behavior. Anyone can join the initiative.
Browse behavioral ratings, neural recordings, and computational tools - all built on the THINGS image set and freely available for research.
24 datasets
No datasets match those filters.
Martin Hebart, Adam Dickter, Alexis KidderMartin Hebart, Adam Dickter, Alexis Kidder, Wan Kwok, Anna Corriveau, Caitlin Van Wicklin, Chris Baker+4 more
National Institute of Mental Health, Bethesda, USA
A freely available database of 26,107 high quality, manually-curated images of 1,854 diverse object concepts, curated systematically from the everyday American English language and using a large-scale web search. Includes 27 high-level categories, semantic embeddings for all concepts, and more metadata.
@article{Hebart2019THINGS,
author = {Hebart, Martin N. and Dickter, Adam H. and Kidder, Alexis and Kwok, Wan Y. and Corriveau, Anna and Van Wicklin, Caitlin and Baker, Chris I.},
title = {{THINGS}: A database of 1,854 object concepts and more than 26,000 naturalistic object images},
journal = {PLOS ONE},
volume = {14},
number = {10},
pages = {e0223792},
year = {2019},
doi = {10.1371/journal.pone.0223792},
url = {https://doi.org/10.1371/journal.pone.0223792}
}
@article{Stoinski2024THINGSplus,
author = {Stoinski, Laura M. and Perkuhn, Jonas and Hebart, Martin N.},
title = {{THINGSplus}: New norms and metadata for the {THINGS} database of 1854 object concepts and 26,107 natural object images},
journal = {Behavior Research Methods},
volume = {56},
number = {3},
pages = {1583--1603},
year = {2024},
doi = {10.3758/s13428-023-02110-8},
url = {https://doi.org/10.3758/s13428-023-02110-8}
}
⚠️ License: Original images are for academic use only. Use THINGSplus license-free images for publications.
Martin Hebart, Oliver Contier, Lina TeichmannMartin Hebart, Oliver Contier, Lina Teichmann, Adam Rockter, Charles Zheng, Alexis Kidder, Anna Corriveau, Maryam Vaziri-Pashkam, Francisco Pereira, Chris Baker+7 more
National Institute of Mental Health, Bethesda, USANational Institute of Mental Health, Bethesda, USA, Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany, Justus Liebig University Giessen, Germany+2 more
More than 4.70 million triplet odd-one-out similarity judgments for 1,854 object images, plus a 66d interpretable embedding. A previous set of 1.46 million triplets served to identify 49 interpretable object dimensions predictive of behavior and similarity (Hebart et al., 2020, Nat Hum Behav).
@article{Hebart2020SPoSE,
author = {Hebart, Martin N. and Zheng, Charles Y. and Pereira, Francisco and Baker, Chris I.},
title = {Revealing the multidimensional mental representations of natural objects underlying human similarity judgements},
journal = {Nature Human Behaviour},
volume = {4},
number = {11},
pages = {1173--1185},
year = {2020},
doi = {10.1038/s41562-020-00951-3},
url = {https://doi.org/10.1038/s41562-020-00951-3}
}
4.70 million triplet judgments from 14,025 participants. "Which object is the odd one out?" responses that reveal similarity structure across all 1,854 concepts.
Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany
New THINGS metadata, with 53 high-level categories, typicality ratings, nameability scores for all images, size ratings, and ratings along several dimensions (e.g. animacy, manipulability, valence, arousal, preciousness, etc.). In addition, 1,854 license-free images were collected that can be used and reproduced (e.g. in publications) without any restriction.
@article{Stoinski2024THINGSplus,
author = {Stoinski, Laura M. and Perkuhn, Jonas and Hebart, Martin N.},
title = {{THINGSplus}: New norms and metadata for the {THINGS} database of 1854 object concepts and 26,107 natural object images},
journal = {Behavior Research Methods},
volume = {56},
number = {3},
pages = {1583--1603},
year = {2024},
doi = {10.3758/s13428-023-02110-8},
url = {https://doi.org/10.3758/s13428-023-02110-8}
}
THINGSplus
About
THINGSplus extends the original THINGS database with comprehensive new norms and metadata for all 1,854 object concepts and 26,107 images.
New metadata includes:
53 high-level categories (expanded from 27)
Typicality ratings for all concepts
Nameability scores for all images
Size ratings
Dimension ratings: animacy, manipulability, valence, arousal, preciousness, and more
THINGSplus also provides license-free alternative images for use in publications.
Download
THINGSplus data is available from the same OSF repository as the original THINGS database:
Size ratings and dimension ratings (animacy, manipulability, valence, arousal, etc.)
Oliver Contier, Martin Hebart, Lina TeichmannOliver Contier, Martin Hebart, Lina Teichmann, Adam Rockter, Charles Zheng, Alexis Kidder, Anna Corriveau, Chris Baker, Maryam Vaziri-Pashkam+6 more
National Institute of Mental Health, Bethesda, USANational Institute of Mental Health, Bethesda, USA, Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany, Justus Liebig University Giessen, Germany+2 more
Event-related functional MRI data in 3 subjects for 8,640 images (720 categories, 12 images per category), collected over the course of 12 sessions. Includes extensive anatomical scans, population receptive field mapping, and functional localizers. Optimized for studying object recognition with a broad and systematic range of object categories.
@article{Hebart2023THINGSdata,
author = {Hebart, Martin N. and Contier, Oliver and Teichmann, Lina and Rockter, Adam H. and Zheng, Charles Y. and Kidder, Alexis and Corriveau, Anna and Vaziri-Pashkam, Maryam and Baker, Chris I.},
title = {{THINGS-data}, a multimodal collection of large-scale datasets for investigating object representations in human brain and behavior},
journal = {eLife},
volume = {12},
pages = {e82580},
year = {2023},
doi = {10.7554/eLife.82580},
url = {https://doi.org/10.7554/eLife.82580}
}
Option 1: DataLad (recommended — download specific subjects)
pip install datalad
datalad clone https://github.com/OpenNeuroDatasets/ds004192.git
cd ds004192
# Download one subject
datalad get sub-01/
# Or download everything
datalad get .
Marie St-Laurent, Basile Pinsard, Oliver ContierMarie St-Laurent, Basile Pinsard, Oliver Contier, Elizabeth DuPre, Katja Seeliger, Valentina Borghesani, Julie A. Boyle, Lune Bellec, Martin N. Hebart+6 more
Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, GermanyMax Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany, Centre de recherche de l’Institut universitaire de gériatrie de Montréal, Montréal, Canada, Département de psychologie, Université de Montréal, Montréal, Canada, Martin Luther University Halle-Wittenberg, Medical Faculty, Halle, Germany, Faculté de psychologie et des sciences de l’éducation, Université de Genève, Genève, Switzerland, Department of Medicine, Justus Liebig University Giessen, Giessen, Germany, Center for Mind, Brain and Behavior (CMBB), Universities of Marburg, Giessen, Darmstadt, Germany+6 more
Densely sampled 3T fMRI, eye-tracking, and behavioral data from four participants completing 33–36 sessions of a continuous recognition task with up to 4,320 images from 720 THINGS categories, each presented three times.
@article{stlaurent2026cneuromodthings,
title = {CNeuroMod-THINGS, a densely-sampled fMRI dataset for visual neuroscience},
author = {St-Laurent, Marie and Pinsard, Basile and Contier, Oliver and DuPre, Elizabeth and Seeliger, Katja and Borghesani, Valentina and Boyle, Julie A. and Bellec, Lune and Hebart, Martin N.},
journal = {Scientific Data},
volume = {13},
pages = {141},
year = {2026},
doi = {10.1038/s41597-026-06591-y},
url = {https://doi.org/10.1038/s41597-026-06591-y}
}
CNeuroMod-THINGS
About
Four participants each completed 33–36 fMRI sessions of a continuous image-recognition task.
Up to 4,320 unique images per participant
720 THINGS object categories
Three presentations of each image
Raw and preprocessed 3T fMRI in BIDS format
Eye tracking, behavior, and physiological recordings
GLMsingle trial-wise and image-wise beta estimates
Functional and retinotopy localizers for three participants
License: CC0. Data files can be retrieved from CONP without registered access.
Download
Install a DataLad version newer than 1.0, then clone the nested dataset:
Download only the submodule and files you need. For example:
cd cneuromod-things/THINGS/glmsingle
datalad get '*'
datalad get -r 'sub-01/qc/*'
Avoid a recursive download of the entire collection. You can alternatively download the version 1.0.1 archive from Zenodo, then use DataLad to retrieve selected files.
Lina Teichmann, Martin Hebart, Oliver ContierLina Teichmann, Martin Hebart, Oliver Contier, Adam Rockter, Charles Zheng, Alexis Kidder, Anna Corriveau, Maryam Vaziri-Pashkam, Chris Baker+6 more
National Institute of Mental Health, Bethesda, USANational Institute of Mental Health, Bethesda, USA, Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany, Justus Liebig University Giessen, Germany+2 more
Magnetoencephalography (MEG) data in 4 subjects for 22,248 images (1,854 categories, 12 images per category), collected over the course of 12 sessions. Optimized for studying object recognition with a broad and systematic range of object categories.
@article{Hebart2023THINGSdata,
author = {Hebart, Martin N. and Contier, Oliver and Teichmann, Lina and Rockter, Adam H. and Zheng, Charles Y. and Kidder, Alexis and Corriveau, Anna and Vaziri-Pashkam, Maryam and Baker, Chris I.},
title = {{THINGS-data}, a multimodal collection of large-scale datasets for investigating object representations in human brain and behavior},
journal = {eLife},
volume = {12},
pages = {e82580},
year = {2023},
doi = {10.7554/eLife.82580},
url = {https://doi.org/10.7554/eLife.82580}
}
Tijl Grootswagers, Ivy Zhou, Amanda RobinsonTijl Grootswagers, Ivy Zhou, Amanda Robinson, Martin Hebart, Thomas Carlson+2 more
MARCS Institute, Western Sydney University, AustraliaMARCS Institute, Western Sydney University, Australia, School of Psychology, University of Sydney, Australia, Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany+2 more
Electroencephalography responses in 50 subjects for 22,248 images (1,854 concepts, 12 images per category), collected in a single session per participant using an RSVP paradigm.
@article{Grootswagers2022THINGSEEG1,
author = {Grootswagers, Tijl and Zhou, Ivy and Robinson, Amanda K. and Hebart, Martin N. and Carlson, Thomas A.},
title = {Human {EEG} recordings for 1,854 concepts presented in rapid serial visual presentation streams},
journal = {Scientific Data},
volume = {9},
number = {1},
pages = {3},
year = {2022},
doi = {10.1038/s41597-021-01102-7},
url = {https://doi.org/10.1038/s41597-021-01102-7}
}
Freie Universität BerlinFreie Universität Berlin, Goethe Universität, Frankfurt am Main+1 more
Raw and preprocessed EEG recordings of 10 participants, each with 82,160 trials spanning 16,740 image conditions coming from the THINGS database.
@article{Gifford2022THINGSEEG2,
author = {Gifford, Alessandro T. and Dwivedi, Kshitij and Roig, Gemma and Cichy, Radoslaw M.},
title = {A large and rich {EEG} dataset for modeling human visual object recognition},
journal = {NeuroImage},
volume = {264},
pages = {119754},
year = {2022},
doi = {10.1016/j.neuroimage.2022.119754},
url = {https://doi.org/10.1016/j.neuroimage.2022.119754}
}
Jonathan Xu, Ugo Bruzadin Nunes, Wangshu JiangJonathan Xu, Ugo Bruzadin Nunes, Wangshu Jiang, Samuel Ryther, Jordan Pringle, Paul S. Scotti, Arnaud Delorme, Reese Kneeland+5 more
AlljoinedAlljoined, University of Waterloo, Sophont, Princeton Neuroscience Institute, University of California San Diego+4 more
Raw and preprocessed 32-channel EEG recordings from 20 participants across four sessions, comprising more than 1.6 million trials for 16,740 unique THINGS images. Collected using consumer-grade EMOTIV FLEX2 wet-electrode hardware and designed for semantic decoding, image retrieval, and EEG-to-image reconstruction.
@misc{xu2025alljoined16mmilliontrialeegimagedataset,
title = {Alljoined-1.6M: A Million-Trial EEG-Image Dataset for Evaluating Affordable Brain-Computer Interfaces},
author = {Jonathan Xu and Ugo Bruzadin Nunes and Wangshu Jiang and Samuel Ryther and Jordan Pringle and Paul S. Scotti and Arnaud Delorme and Reese Kneeland},
year = {2025},
eprint = {2508.18571},
archivePrefix = {arXiv},
primaryClass = {q-bio.NC},
url = {https://arxiv.org/abs/2508.18571}
}
Alljoined-1.6M (EEG)
136 GB
About
20 participants × 4 sessions
More than 1.6 million trials
16,740 unique THINGS images
32-channel EEG sampled at 256 Hz
What's Included
raw_eeg/ — per-participant and per-session recordings
preprocessed_eeg/ — NumPy arrays and Parquet metadata
stimuli.zip — stimulus images
Questionnaires and a stimulus-presentation video
Size: approximately 136 GB
License confirmation pending. Until it is resolved, consult the license and additional terms published with the dataset on Hugging Face.
Card image: unmodified Figure 1 by Jonathan Xu et al., copied from arXiv v2 under CC BY-NC-ND 4.0.
Download
Install the Hugging Face client:
pip install -U huggingface_hub
Download a local snapshot:
from huggingface_hub import snapshot_download
snapshot_download(
repo_id="Alljoined/Alljoined-1.6M",
repo_type="dataset",
)
Dept of Vision & Cognition, Netherlands Institute for Neuroscience, Amsterdam, The Netherlands
High-channel count electrophysiological recordings of 22,248 images (1,854 categories, 12 images per category) from the macaque visual cortex across V1, V4, and IT, in two animals.
@article{Papale2025TVSD,
author = {Papale, Paolo and Wang, Feng and Self, Matthew W. and Roelfsema, Pieter R.},
title = {An extensive dataset of spiking activity to reveal the syntax of the ventral stream},
journal = {Neuron},
volume = {113},
number = {4},
pages = {539--553.e5},
year = {2025},
doi = {10.1016/j.neuron.2024.12.003},
url = {https://doi.org/10.1016/j.neuron.2024.12.003}
}
Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany
Streamlines the extraction of neural network activations by providing a simple wrapper for extracting activations from a wide range of commonly used deep convolutional neural network architectures.
@article{Muttenthaler2021THINGSvision,
author = {Muttenthaler, Lukas and Hebart, Martin N.},
title = {{THINGSvision}: A {Python} Toolbox for Streamlining the Extraction of Activations From Deep Neural Networks},
journal = {Frontiers in Neuroinformatics},
volume = {15},
pages = {679838},
year = {2021},
doi = {10.3389/fninf.2021.679838},
url = {https://doi.org/10.3389/fninf.2021.679838}
}
Python package to extract activations from 100+ neural network models including torchvision, timm, CLIP, OpenCLIP, DINO, and more.
Features
Unified API across model families
Batch processing for large datasets
Built-in RSA and CKA analysis
GPU acceleration
Download
Install from PyPI:
pip install thingsvision
Extract features using the Python API:
from thingsvision import get_extractor
extractor = get_extractor(
model_name="alexnet",
source="torchvision",
device="cuda",
pretrained=True,
)
features = extractor.extract_features(
batches=dataloader,
module_name="features.10",
)
See the documentation for CLI usage and supported models (CLIP, DINO, DINOv2, etc.).
Max Kramer, Martin Hebart, Chris BakerMax Kramer, Martin Hebart, Chris Baker, Wilma Bainbridge+1 more
Department of Psychology, University of Chicago, USADepartment of Psychology, University of Chicago, USA, Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany, National Institute of Mental Health, Bethesda, USA+2 more
Memorability scores for all 26,107 object images, collected in a large sample of >13,000 participants. Offers a systematic evaluation of memorability across a wide range of natural object images, object concepts, and high-level categories.
@article{Kramer2023Memorability,
author = {Kramer, Max A. and Hebart, Martin N. and Baker, Chris I. and Bainbridge, Wilma A.},
title = {The features underlying the memorability of objects},
journal = {Science Advances},
volume = {9},
number = {17},
pages = {eadd2981},
year = {2023},
doi = {10.1126/sciadv.add2981},
url = {https://doi.org/10.1126/sciadv.add2981}
}
Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany
Semantic feature production norms for all 1,854 object concepts in THINGS, generated with the natural language model GPT-3, developed by OpenAI.
@article{Hansen2022SemanticFeatures,
author = {Hansen, Hannes and Hebart, Martin N.},
title = {Semantic features of object concepts generated with {GPT-3}},
journal = {arXiv preprint},
year = {2022},
eprint = {2202.03753},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2202.03753}
}
THINGS semantic feature norm
About
This dataset provides semantic feature norms for the 1,854 object concepts in the THINGS database, automatically generated using GPT-3.
Key features:
Semantic features for all 1,854 THINGS concepts
Generated using GPT-3 language model
Comparable to human-generated feature norms
Useful for studying conceptual representations
The generated features rival human norms in predicting similarity, relatedness, and category membership.
Download
Feature norms are available from the OSF repository:
pip install osfclient
osf -p jum2f clone
Look for the semantic feature files in the downloaded repository.
Jaan Aru, Kadi Tulver, Tarun Khajuria
University of Tartu
A dataset of "constellation" images that can be used to study inference in human vision and AI. The images are stripped of local details creating a dotted outline of the object that can be inferred from the local pattern. The dataset includes 3533 image sets of a total of 1215 common objects from the THINGS dataset. A selected set of 481 top constellation images and the code to generate more constellation images from photos are also included.
@inproceedings{Khajuria2022Constellations,
author = {Khajuria, Tarun and Hebart, Martin N. and Battleday, Ruairidh M.},
title = {Constellations: A Novel Dataset for Studying Iterative Inference in Humans and {AI}},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
year = {2022},
pages = {4392--4400},
url = {https://openaccess.thecvf.com/content/CVPR2022W/SketchDL/papers/Khajuria_Constellations_A_Novel_Dataset_for_Studying_Iterative_Inference_in_Humans_CVPRW_2022_paper.pdf}
}
3,533 constellation images from 1,215 objects — dotted outline representations for studying visual inference and object recognition from minimal information.
Filipp Schmidt, Martin Hebart, Alex SchmidFilipp Schmidt, Martin Hebart, Alex Schmid, Roland Fleming+1 more
Psychology Department, University Gießen, GermanyPsychology Department, University Gießen, Germany, Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany, National Institute of Mental Health, Bethesda, USA+2 more
600 images of 200 materials sampled systematically and representatively from the American English language. Includes object dimensions and material similarity matrices identified from >1.8 million similarity judgments.
@article{Schmidt2025STUFF,
author = {Schmidt, Filipp and Hebart, Martin N.},
title = {Core dimensions of human material perception},
journal = {Proceedings of the National Academy of Sciences},
volume = {122},
number = {10},
pages = {e2417202122},
year = {2025},
doi = {10.1073/pnas.2417202122},
url = {https://doi.org/10.1073/pnas.2417202122}
}
200 material concepts with 600 images (3 per category) and 1.87 million similarity judgments. Companion to THINGS for material/texture perception research.
Kushin Mukherjee, Holly Huey, Laura M. StoinskiKushin Mukherjee, Holly Huey, Laura M. Stoinski, Martin N. Hebart, Judith E. Fan, Wilma A. Bainbridge+3 more
Department of Psychology, University of Wisconsin-Madison, Madison, WI, USADepartment of Psychology, University of Wisconsin-Madison, Madison, WI, USA, Wisconsin Institute for Discovery, University of Wisconsin-Madison, Madison, WI, USA, Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany, Department of Psychology, University of California, San Diego, CA, USA, University of Leipzig, Leipzig, Germany, International Max Planck Research School on Cognitive NeuroImaging, Leipzig, Germany, Department of Medicine, Justus Liebig University, Giessen, Germany, Center for Mind, Brain and Behavior, Universities of Marburg, Giessen, and Darmstadt, Marburg, Germany, Department of Psychology, Stanford University, Stanford, CA, USA, Department of Psychology, University of Chicago, Chicago, IL, USA, Neuroscience Institute, University of Chicago, Chicago, IL, USA+10 more
A large-scale collection of 28,627 human drawings spanning all 1,854 THINGS object concepts, with stroke histories, drawing-level recognizability, image recognizability, and participant metadata.
@article{mukherjee2026drawingsofthings,
title = {Drawings of THINGS: A large-scale drawing dataset of 1854 object concepts},
author = {Mukherjee, Kushin and Huey, Holly and Stoinski, Laura M. and Hebart, Martin N. and Fan, Judith E. and Bainbridge, Wilma A.},
journal = {Behavior Research Methods},
volume = {58},
pages = {57},
year = {2026},
doi = {10.3758/s13428-025-02887-w},
url = {https://doi.org/10.3758/s13428-025-02887-w}
}
Drawings of THINGS
About
28,627 validated human drawings covering all 1,854 THINGS object concepts, collected from 1,314 participants.
550 × 550 pixel raster drawings
Stroke histories and drawing timing
Drawing-level recognizability scores
Photo recognizability ratings for THINGS images
Participant demographics, drawing skill, and mental imagery measures
The dataset mirrors the concept structure of the THINGS image database and supports comparisons between drawings, photographs, and semantic representations.
Card image: unmodified Figure 1 by Kushin Mukherjee et al., copied from the published article under CC BY 4.0.
Analysis code and documentation are available in the linked GitHub repository.
Philipp A. Schumann, Rico Stecher, Martin N. HebartPhilipp A. Schumann, Rico Stecher, Martin N. Hebart, Daniel Kaiser+1 more
Neural Computation Group, Justus Liebig University Giessen, Giessen, GermanyNeural Computation Group, Justus Liebig University Giessen, Giessen, Germany, Department of Psychology, Carl von Ossietzky University Oldenburg, Oldenburg, Germany, Institute of Systems Neuroscience, University Medical Center Hamburg-Eppendorf, Hamburg, Germany, Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany, Department of Medicine, Justus Liebig University Giessen, Giessen, Germany, Center for Mind, Brain and Behavior, Philipps University Marburg, Justus Liebig University Giessen and Technical University Darmstadt, Marburg, Germany, Cluster of Excellence The Adaptive Mind, Justus Liebig University Giessen, Philipps University Marburg and Technical University Darmstadt, Giessen, Germany, Center for Applied Computer Science and Data Science, Justus Liebig University Giessen, Giessen, Germany+7 more
Beauty ratings for 12,978 THINGS images spanning all 1,854 object concepts, collected from 3,750 observers and averaged across raters and seven exemplars per concept.
@article{schumann2026beauty,
title = {Primate visual cortex spontaneously computes the beauty of objects},
author = {Schumann, Philipp A. and Stecher, Rico and Hebart, Martin N. and Kaiser, Daniel},
journal = {bioRxiv},
year = {2026},
doi = {10.64898/2026.07.07.736498},
url = {https://doi.org/10.64898/2026.07.07.736498}
}
THINGS-beauty
About
3,750 online observers rated the beauty of 12,978 THINGS images spanning all 1,854 object concepts.
Seven image exemplars per object concept
24–36 ratings per image (30.51 on average)
Image- and concept-level beauty ratings
Associated object-property and cuteness ratings
Representational-similarity data linking a 634-concept subset to EEG and fMRI
The public Zenodo release accompanies the bioRxiv preprint.
Card image: Figure 1 by Philipp A. Schumann et al., copied from the bioRxiv preprint under CC BY 4.0; cropped to the overview panels.
Download
Download the core rating files individually from Zenodo:
Browse the Zenodo record to download the larger EEG, MEG, fMRI, fusion, and analysis files separately.
Changde Du, Kaicheng Fu, Bincheng WenChangde Du, Kaicheng Fu, Bincheng Wen, Yi Sun, Jie Peng, Wei Wei, Ying Gao, Shengpei Wang, Chuncheng Zhang, Jinpeng Li, Shuang Qiu, Le Chang, Huiguang He+10 more
State Key Laboratory of Brain Cognition and Brain-inspired Intelligence Technology, Institute of Automation, Chinese Academy of Sciences, Beijing, ChinaState Key Laboratory of Brain Cognition and Brain-inspired Intelligence Technology, Institute of Automation, Chinese Academy of Sciences, Beijing, China, School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing, China, Institute of Neuroscience, CAS Center for Excellence in Brain Science and Intelligence Technology, Chinese Academy of Sciences, Shanghai, China, School of Automation Science and Engineering, South China University of Technology, Guangzhou, China, Zhongguancun Academy, Beijing, China+4 more
A 66-dimensional embedding of all 1,854 THINGS object concepts, derived from 4.7 million triplet judgments generated with language-only and multimodal large language models and validated against human behavior.
@article{du2025humanlike,
title = {Human-like object concept representations emerge naturally in multimodal large language models},
author = {Du, Changde and Fu, Kaicheng and Wen, Bincheng and Sun, Yi and Peng, Jie and Wei, Wei and Gao, Ying and Wang, Shengpei and Zhang, Chuncheng and Li, Jinpeng and Qiu, Shuang and Chang, Le and He, Huiguang},
journal = {Nature Machine Intelligence},
volume = {7},
pages = {860--875},
year = {2025},
doi = {10.1038/s42256-025-01049-z},
url = {https://doi.org/10.1038/s42256-025-01049-z}
}
THINGS-LLM object representations
About
4.7 million triplet judgments characterize the similarity structure of all 1,854 THINGS object concepts.
A sparse positive similarity embedding with 66 interpretable dimensions
Representations derived from language-only and multimodal large language models
Comparison with human behavioral judgments and neural representations
Concept labels and analysis resources for reproducing the published results
The data are hosted on OSF and the analysis code is available on GitHub.
Card image: unmodified Figure 1 by Changde Du et al., copied from arXiv v3 under CC BY-NC-ND 4.0.
Download
Browse and download the public data files from the OSF project.
For a command-line download, install osfclient and clone the project:
Analysis code and usage details are available in the linked GitHub repository.
Thomas Reber, Florian Mormann
Dept of Epileptology, University of Bonn
Direct recordings from human entorhinal cortex, hippocampus, amygdala, and parahippocampal cortex in 23 patients for 1,200 images (150 categories, 8 images each).
THINGS electrophysiology1
data analysis ongoing
About
Human electrophysiology recordings in response to THINGS images, providing high temporal resolution neural data.
Siegel Lab
University of Tübingen, Germany
Scalp EEG recordings of 8,640 THINGS images (720 categories, 12 images per category) in 2 macaque monkeys.
THINGS macaque EEG
data analysis ongoing
About
EEG recordings from macaque monkeys viewing THINGS images, enabling cross-species comparisons of object representations.
Ratan Murty, Sachi Sanghavi
Massachusetts Institute of Technology, Cambridge MA, USA
Electrophysiological recordings of 14,832 THINGS images (1,854 categories, 8 images per category) in area V4 of a macaque monkey.
THINGS macaque V4
data collection completed
About
Neural recordings from macaque area V4 in response to THINGS images, probing mid-level visual representations.
Avniel Ghuman, LCND team
Laboratory of Cognitive Neurodynamics, University of Pittsburgh, PA, USA
Intracranial EEG from ventral temporal cortex in human patients.
THINGS iEEG
data collection ongoing
About
Intracranial EEG recordings from human patients viewing THINGS images, providing unique spatiotemporal resolution of object processing.
PRISME team
Institut Universitaire en Santé Mentale de Montréal (IUSMM), Canada
Event-related functional MRI measurements at 3T in a large sample of psychotic patients, presenting 5,568 images from 720 object categories per patient.
THINGS fMRI3
data collection ongoing
About
An fMRI dataset collected as part of an international collaboration extending THINGS neural data collection.
Deniz Vatansever
Zhangjiang International Brain Imaging Centre, Fudan University, Shanghai, China
Dense-sampling object-recognition fMRI from 20 participants across five 7T sessions, acquired at 1.5 mm isotropic resolution using images from all 1,854 THINGS concepts.
THINGS-fMRI-7T
Coming soon
About
20 participants completed five 7T sessions of a continuous object-recognition task.
60 object-recognition runs per participant (12 per session)
1.5 mm isotropic task fMRI
1,280 images and 3,840 image trials per participant
1,104 participant-specific images plus 176 images shared across participants
Each image repeated three times within participant
22,256 THINGS images and all 1,854 concepts represented across the experiment
Functional-localizer, behavioral, and auxiliary assessment data
The dataset and accompanying manuscript are in preparation for public release.
Card image: temporary THINGS stimulus collage copied from the existing THINGS concepts and images card.
Join the initiative
The THINGS initiative is an open science project. We invite researchers to use these datasets, contribute new modalities, or expand the database.
