{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"institution":[{"name":"bioRxiv"}],"indexed":{"date-parts":[[2026,10,29]],"date-time":"2026-10-29T18:42:32Z","timestamp":1793299352686,"version":"4.5.0"},"posted":{"date-parts":[[2026,8,6]]},"group-title":"Neuroscience","reference-count":81,"publisher":"openRxiv","license":[{"start":{"date-parts":[[2026,8,6]],"date-time":"2026-08-06T00:00:00Z","timestamp":1785974400000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/www.biorxiv.org\/about\/FAQ#license"}],"funder":[{"id":[{"id":"https:\/\/ror.org\/000dswa46","id-type":"ROR","asserted-by":"publisher"}]},{"DOI":"10.13039\/100000002","name":"National Institutes of Health","doi-asserted-by":"crossref","award":["R01EY032085"],"award-info":[{"award-number":["R01EY032085"]}],"id":[{"id":"10.13039\/100000002","id-type":"DOI","asserted-by":"crossref"}]},{"id":[{"id":"https:\/\/ror.org\/00gc20a07","id-type":"ROR","asserted-by":"publisher"}]},{"id":[{"id":"https:\/\/ror.org\/01cjash87","id-type":"ROR","asserted-by":"publisher"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"accepted":{"date-parts":[[2026,8,6]]},"abstract":"<jats:title>SUMMARY<\/jats:title>\n                <jats:p>Deep neural networks (DNNs) are promising computational models for understanding visual object recognition. Yet, whether DNNs use similar visual cues for object recognition as humans do remains unknown. We created an image set that systematically untangles global shape, internal parts, and texture information, and compared human recognition behavior against &gt;200 DNNs spanning diverse architectures, training diets, and training objectives. No DNNs replicated humans\u2019 cue-reliance profile, including those with recurrence or specialized training. Fine-tuned text-image contrastive-trained models, regardless of architecture, were most human-like overall, but lost their human-alignment when the global shape was disrupted. Strikingly, all DNNs substantially underperformed humans when the global shape cue alone was critical to object recognition. Furthermore, alignment with ventral stream neural recordings in an existing database did not predict alignment to human behavior, and model performance does not always predict its human-alignment. Together, these findings reveal systematic and persistent differences between human and machine vision.<\/jats:p>","DOI":"10.64898\/2026.08.01.742229","type":"posted-content","created":{"date-parts":[[2026,8,7]],"date-time":"2026-08-07T04:42:14Z","timestamp":1786077734000},"source":"Crossref","is-referenced-by-count":2,"title":["Systematic image perturbations reveal persistent gaps between human and machine vision"],"prefix":"10.64898","author":[{"ORCID":"https:\/\/orcid.org\/0000-0003-3239-0878","authenticated-orcid":false,"given":"Mugihiko","family":"Kato","sequence":"first","affiliation":[{"name":"Department of Neuroscience, New York University Grossman School of Medicine, New York, NY 10016, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-1549-1351","authenticated-orcid":false,"given":"Biyu J.","family":"He","sequence":"additional","affiliation":[{"name":"Department of Neuroscience, New York University Grossman School of Medicine, New York, NY 10016, USA"},{"name":"Department of Neurology, New York University Grossman School of Medicine, New York, NY 10016, USA"},{"name":"Department of Radiology, New York University Grossman School of Medicine, New York, NY 10016, USA"},{"name":"Department of Biomedical Engineering, NYU Tandon School of Engineering, New York, NY 11201, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"54368","reference":[{"key":"2026080902551145000_2026.08.01.742229v1.1","doi-asserted-by":"crossref","unstructured":"Kar, K. , and DiCarlo, J.J . (2024). The quest for an integrated set of neural mechanisms underlying object recognition in primates. Annual Review of Vision Science 10.","DOI":"10.1146\/annurev-vision-112823-030616"},{"key":"2026080902551145000_2026.08.01.742229v1.2","doi-asserted-by":"publisher","DOI":"10.1186\/s40537-021-00444-8"},{"key":"2026080902551145000_2026.08.01.742229v1.3","doi-asserted-by":"crossref","unstructured":"He, K. , Zhang, X. , Ren, S. , and Sun, J . (2015). Delving deep into rectifiers: Surpassing human-level performance on imagenet classification. pp. 1026\u20131034.","DOI":"10.1109\/ICCV.2015.123"},{"key":"2026080902551145000_2026.08.01.742229v1.4","unstructured":"Shankar, V. , Roelofs, R. , Mania, H. , Fang, A. , Recht, B. , and Schmidt, L. (2020). Evaluating Machine Accuracy on ImageNet. In D. Hal, III , and S. Aarti , eds. Proceedings of the 37th International Conference on Machine Learning. PMLR."},{"key":"2026080902551145000_2026.08.01.742229v1.5","doi-asserted-by":"publisher","DOI":"10.1038\/srep27755"},{"key":"2026080902551145000_2026.08.01.742229v1.6","doi-asserted-by":"publisher","DOI":"10.1523\/JNEUROSCI.5023-14.2015"},{"key":"2026080902551145000_2026.08.01.742229v1.7","doi-asserted-by":"publisher","DOI":"10.1038\/ncomms15037"},{"key":"2026080902551145000_2026.08.01.742229v1.8","doi-asserted-by":"publisher","DOI":"10.1073\/pnas.1403112111"},{"key":"2026080902551145000_2026.08.01.742229v1.9","doi-asserted-by":"publisher","DOI":"10.1038\/nn.4244"},{"key":"2026080902551145000_2026.08.01.742229v1.10","doi-asserted-by":"publisher","DOI":"10.1038\/s41467-024-53147-y"},{"key":"2026080902551145000_2026.08.01.742229v1.11","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2106.07411"},{"key":"2026080902551145000_2026.08.01.742229v1.12","unstructured":"Szegedy, C. , Zaremba, W. , Sutskever, I. , Bruna, J. , Erhan, D. , Goodfellow, I. , and Fergus, R. (2013). Intriguing properties of neural networks. arXiv preprint arXiv:1312.6199."},{"key":"2026080902551145000_2026.08.01.742229v1.13","unstructured":"Geirhos, R. , Temme, C.R. , Rauber, J. , Sch\u00fctt, H.H. , Bethge, M. , and Wichmann, F.A . (2018). Generalisation in humans and deep neural networks. Advances in neural information processing systems 31."},{"key":"2026080902551145000_2026.08.01.742229v1.14","doi-asserted-by":"crossref","unstructured":"Dodge, S. , and Karam, L . (2017). A study and comparison of human and deep learning recognition performance under visual distortions. 2017 26th international conference on computer communication and networks (ICCCN), 1\u20137.","DOI":"10.1109\/ICCCN.2017.8038465"},{"key":"2026080902551145000_2026.08.01.742229v1.15","doi-asserted-by":"publisher","DOI":"10.1371\/journal.pbio.3001418"},{"key":"2026080902551145000_2026.08.01.742229v1.16","doi-asserted-by":"publisher","DOI":"10.1167\/jov.21.5.6"},{"key":"2026080902551145000_2026.08.01.742229v1.17","unstructured":"Muttenthaler, L. , Dippel, J. , Linhardt, L. , Vandermeulen, R.A. , and Kornblith, S. (2022). Human alignment of neural network representations. arXiv preprint arXiv:2211.01201."},{"key":"2026080902551145000_2026.08.01.742229v1.18","doi-asserted-by":"publisher","DOI":"10.1016\/j.isci.2021.103013"},{"key":"2026080902551145000_2026.08.01.742229v1.19","unstructured":"Gokce, A. , and Schrimpf, M. (2024). Scaling laws for task-optimized models of the primate visual ventral stream. arXiv preprint arXiv:2411.05712."},{"key":"2026080902551145000_2026.08.01.742229v1.20","unstructured":"Geirhos, R. , Rubisch, P. , Michaelis, C. , Bethge, M. , Wichmann, F.A. , and Brendel, W. (2018). ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness. International conference on learning representations."},{"key":"2026080902551145000_2026.08.01.742229v1.21","doi-asserted-by":"publisher","DOI":"10.1371\/journal.pcbi.1006613"},{"key":"2026080902551145000_2026.08.01.742229v1.22","doi-asserted-by":"publisher","DOI":"10.1016\/j.conb.2017.08.019"},{"key":"2026080902551145000_2026.08.01.742229v1.23","doi-asserted-by":"publisher","DOI":"10.1073\/pnas.2115302119"},{"key":"2026080902551145000_2026.08.01.742229v1.24","unstructured":"Brendel, W. , and Bethge, M. (2019). Approximating cnns with bag-of-local-features models works surprisingly well on imagenet. arXiv preprint arXiv:1904.00760."},{"key":"2026080902551145000_2026.08.01.742229v1.25","doi-asserted-by":"publisher","DOI":"10.1016\/j.isci.2022.104913"},{"key":"2026080902551145000_2026.08.01.742229v1.26","doi-asserted-by":"publisher","DOI":"10.1038\/s41467-021-22078-3"},{"key":"2026080902551145000_2026.08.01.742229v1.27","doi-asserted-by":"publisher","DOI":"10.1016\/j.visres.2020.04.003"},{"key":"2026080902551145000_2026.08.01.742229v1.28","first-page":"05253","article-title":"Contour Integration Underlies Human-Like Vision","volume":"2504","year":"2025","journal-title":"arXiv preprint arXiv"},{"key":"2026080902551145000_2026.08.01.742229v1.29","unstructured":"Jaini, P. , Clark, K. , and Geirhos, R. (2023). Intriguing properties of generative classifiers. arXiv preprint arXiv:2309.16779."},{"key":"2026080902551145000_2026.08.01.742229v1.30","doi-asserted-by":"crossref","first-page":"1415","DOI":"10.1038\/s42256-023-00753-y","article-title":"Better models of human high-level visual cortex emerge from natural language supervision with a large and diverse dataset","volume":"5","year":"2023","journal-title":"Nature Machine Intelligence"},{"key":"2026080902551145000_2026.08.01.742229v1.31","doi-asserted-by":"crossref","unstructured":"Du, C. , Fu, K. , Wen, B. , Sun, Y. , Peng, J. , Wei, W. , Gao, Y. , Wang, S. , Zhang, C. , and Li, J . (2025). Human-like object concept representations emerge naturally in multimodal large language models. Nature Machine Intelligence, 1-16.","DOI":"10.1038\/s42256-025-01049-z"},{"key":"2026080902551145000_2026.08.01.742229v1.32","doi-asserted-by":"publisher","DOI":"10.1038\/s41467-024-45679-0"},{"key":"2026080902551145000_2026.08.01.742229v1.33","doi-asserted-by":"crossref","unstructured":"Kubilius, J. , Schrimpf, M. , Nayebi, A. , Bear, D. , Yamins, D.L. , and DiCarlo, J.J. (2018). Cornet: Modeling the neural mechanisms of core object recognition. BioRxiv, 408385.","DOI":"10.1101\/408385"},{"key":"2026080902551145000_2026.08.01.742229v1.34","doi-asserted-by":"publisher","DOI":"10.3389\/fpsyg.2017.01551"},{"key":"2026080902551145000_2026.08.01.742229v1.35","doi-asserted-by":"publisher","DOI":"10.1145\/219717.219748"},{"key":"2026080902551145000_2026.08.01.742229v1.36","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2006.16736"},{"key":"2026080902551145000_2026.08.01.742229v1.37","unstructured":"Kubilius, J. , Schrimpf, M. , Kar, K. , Rajalingham, R. , Hong, H. , Majaj, N. , Issa, E. , Bashivan, P. , Prescott-Roy, J. , and Schmidt, K . (2019). Brain-like object recognition with high-performing shallow recurrent ANNs. Advances in neural information processing systems 32."},{"key":"2026080902551145000_2026.08.01.742229v1.38","doi-asserted-by":"publisher","DOI":"10.1038\/s41593-019-0392-5"},{"key":"2026080902551145000_2026.08.01.742229v1.39","doi-asserted-by":"publisher","DOI":"10.1073\/pnas.1719397115"},{"key":"2026080902551145000_2026.08.01.742229v1.40","doi-asserted-by":"publisher","DOI":"10.1371\/journal.pcbi.1007001"},{"key":"2026080902551145000_2026.08.01.742229v1.41","doi-asserted-by":"publisher","DOI":"10.1016\/j.neuron.2020.09.035"},{"key":"2026080902551145000_2026.08.01.742229v1.42","first-page":"11929","article-title":"An image is worth 16\u00d716 words: Transformers for image recognition at scale","volume":"2010","year":"2020","journal-title":"arXiv preprint arXiv"},{"key":"2026080902551145000_2026.08.01.742229v1.43","unstructured":"Dong, X. , Bao, J. , Zhang, T. , Chen, D. , Gu, S. , Zhang, W. , Yuan, L. , Chen, D. , Wen, F. , and Yu, N. (2022). Clip itself is a strong fine-tuner: Achieving 85.7% and 88.0% top-1 accuracy with vit-b and vit-l on imagenet. arXiv preprint arXiv:2212.06138."},{"key":"2026080902551145000_2026.08.01.742229v1.44","doi-asserted-by":"crossref","unstructured":"Schrimpf, M. , Kubilius, J. , Hong, H. , Majaj, N.J. , Rajalingham, R. , Issa, E.B. , Kar, K. , Bashivan, P. , Prescott-Roy, J. , and Geiger, F . (2018). Brain-score: Which artificial neural network for object recognition is most brain-like? BioRxiv, 407007.","DOI":"10.1101\/407007"},{"key":"2026080902551145000_2026.08.01.742229v1.45","doi-asserted-by":"publisher","DOI":"10.1016\/j.neuron.2020.07.040"},{"key":"2026080902551145000_2026.08.01.742229v1.46","doi-asserted-by":"publisher","DOI":"10.1016\/j.neuron.2012.01.010"},{"key":"2026080902551145000_2026.08.01.742229v1.47","unstructured":"Hendrycks, D. , and Dietterich, T. (2019). Benchmarking neural network robustness to common corruptions and perturbations. arXiv preprint arXiv:1903.12261."},{"key":"2026080902551145000_2026.08.01.742229v1.48","unstructured":"Ahlert, J. , Klein, T. , Wichmann, F. , and Geirhos, R. (2024). How aligned are different alignment metrics? arXiv preprint arXiv:2407.07530."},{"key":"2026080902551145000_2026.08.01.742229v1.49","doi-asserted-by":"publisher","DOI":"10.7554\/eLife.27576"},{"key":"2026080902551145000_2026.08.01.742229v1.50","doi-asserted-by":"publisher","DOI":"10.1073\/pnas.0507062103"},{"key":"2026080902551145000_2026.08.01.742229v1.51","doi-asserted-by":"publisher","DOI":"10.7554\/eLife.25784"},{"key":"2026080902551145000_2026.08.01.742229v1.52","doi-asserted-by":"publisher","DOI":"10.1073\/pnas.1221945110"},{"key":"2026080902551145000_2026.08.01.742229v1.53","doi-asserted-by":"publisher","DOI":"10.1523\/JNEUROSCI.2314-15.2016"},{"key":"2026080902551145000_2026.08.01.742229v1.54","doi-asserted-by":"publisher","DOI":"10.1038\/s41467-021-23266-x"},{"key":"2026080902551145000_2026.08.01.742229v1.55","doi-asserted-by":"publisher","DOI":"10.1093\/texcom\/tgad003"},{"key":"2026080902551145000_2026.08.01.742229v1.56","doi-asserted-by":"publisher","DOI":"10.1038\/nn2036"},{"key":"2026080902551145000_2026.08.01.742229v1.57","doi-asserted-by":"crossref","unstructured":"Doerig, A. , Kietzmann, T.C. , Allen, E. , Wu, Y. , Naselaris, T. , Kay, K. , and Charest, I . (2025). High-level visual representations in the human brain are aligned with large language models. Nature Machine Intelligence, 1\u201315.","DOI":"10.1038\/s42256-025-01072-0"},{"key":"2026080902551145000_2026.08.01.742229v1.58","doi-asserted-by":"publisher","DOI":"10.3758\/s13414-013-0605-z"},{"key":"2026080902551145000_2026.08.01.742229v1.59","doi-asserted-by":"publisher","DOI":"10.1016\/0010-0285(76)90013-X"},{"key":"2026080902551145000_2026.08.01.742229v1.60","doi-asserted-by":"publisher","DOI":"10.1073\/pnas.2011417118"},{"key":"2026080902551145000_2026.08.01.742229v1.61","doi-asserted-by":"publisher","DOI":"10.1038\/s41586-025-09631-6"},{"key":"2026080902551145000_2026.08.01.742229v1.62","doi-asserted-by":"publisher","DOI":"10.1073\/pnas.2417202122"},{"key":"2026080902551145000_2026.08.01.742229v1.63","doi-asserted-by":"publisher","DOI":"10.1523\/JNEUROSCI.1822-22.2023"},{"key":"2026080902551145000_2026.08.01.742229v1.64","doi-asserted-by":"crossref","unstructured":"Casile, A. , Cordier, A. , Kim, J.G. , Cometa, A. , Madsen, J.R. , Stone, S. , Ben-Yosef, G. , Ullman, S. , Anderson, W. , and Kreiman, G . (2025). Neural correlates of minimal recognizable configurations in the human brain. Cell reports 44.","DOI":"10.1016\/j.celrep.2025.115429"},{"key":"2026080902551145000_2026.08.01.742229v1.65","doi-asserted-by":"publisher","DOI":"10.1016\/j.tics.2022.09.019"},{"key":"2026080902551145000_2026.08.01.742229v1.66","doi-asserted-by":"crossref","unstructured":"Cherian, T. , Jacob, G. , and Arun, S. (2025). Do monkeys see the way we do? Qualitative similarities and differences between monkey and human perception. bioRxiv, 2025.2001. 2030.635614.","DOI":"10.1101\/2025.01.30.635614"},{"key":"2026080902551145000_2026.08.01.742229v1.67","doi-asserted-by":"publisher","DOI":"10.1007\/s11263-015-0816-y"},{"key":"2026080902551145000_2026.08.01.742229v1.68","doi-asserted-by":"publisher","DOI":"10.1371\/journal.pcbi.1004896"},{"key":"2026080902551145000_2026.08.01.742229v1.69","unstructured":"Gatys, L. , Ecker, A.S. , and Bethge, M . (2015). Texture synthesis using convolutional neural networks. Advances in neural information processing systems 28."},{"key":"2026080902551145000_2026.08.01.742229v1.70","doi-asserted-by":"crossref","unstructured":"Gatys, L.A. , Ecker, A.S. , and Bethge, M . (2016). Image style transfer using convolutional neural networks. pp. 2414\u20132423.","DOI":"10.1109\/CVPR.2016.265"},{"key":"2026080902551145000_2026.08.01.742229v1.71","doi-asserted-by":"crossref","unstructured":"Chan, C. , Durand, F. , and Isola, P . (2022). Learning to generate line drawings that convey geometry and semantics. pp. 7915\u20137925.","DOI":"10.1109\/CVPR52688.2022.00776"},{"key":"2026080902551145000_2026.08.01.742229v1.72","doi-asserted-by":"publisher","DOI":"10.1523\/JNEUROSCI.1546-22.2022"},{"key":"2026080902551145000_2026.08.01.742229v1.73","doi-asserted-by":"publisher","DOI":"10.7554\/eLife.36068"},{"key":"2026080902551145000_2026.08.01.742229v1.74","unstructured":"Paszke, A. (2019). Pytorch: An imperative style, high-performance deep learning library. arXiv preprint arXiv:1912.01703."},{"key":"2026080902551145000_2026.08.01.742229v1.75","doi-asserted-by":"crossref","unstructured":"Liu, Z. , Mao, H. , Wu, C.-Y. , Feichtenhofer, C. , Darrell, T. , and Xie, S . (2022). A convnet for the 2020s. pp. 11976\u201311986.","DOI":"10.1109\/CVPR52688.2022.01167"},{"key":"2026080902551145000_2026.08.01.742229v1.76","first-page":"25278","article-title":"Laion-5B: An open large-scale dataset for training next generation image-text models","volume":"35","year":"2022","journal-title":"Advances in neural information processing systems"},{"key":"2026080902551145000_2026.08.01.742229v1.77","doi-asserted-by":"publisher","DOI":"10.5281\/zenodo.5143772"},{"key":"2026080902551145000_2026.08.01.742229v1.78","doi-asserted-by":"publisher","DOI":"10.5281\/zenodo.4414861"},{"key":"2026080902551145000_2026.08.01.742229v1.79","doi-asserted-by":"crossref","unstructured":"Li, A.C. , Prabhudesai, M. , Duggal, S. , Brown, E. , and Pathak, D . (2023). Your diffusion model is secretly a zero-shot classifier. pp. 2206\u20132217.","DOI":"10.1109\/ICCV51070.2023.00210"},{"key":"2026080902551145000_2026.08.01.742229v1.80","doi-asserted-by":"publisher","DOI":"10.1523\/JNEUROSCI.0388-18.2018"},{"key":"2026080902551145000_2026.08.01.742229v1.81","doi-asserted-by":"publisher","DOI":"10.1038\/s42256-025-01041-7"}],"container-title":[],"original-title":[],"link":[{"URL":"https:\/\/syndication.highwire.org\/content\/doi\/10.64898\/2026.08.01.742229","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,8,9]],"date-time":"2026-08-09T09:55:24Z","timestamp":1786269324000},"score":1,"resource":{"primary":{"URL":"http:\/\/biorxiv.org\/lookup\/doi\/10.64898\/2026.08.01.742229"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,8,6]]},"references-count":81,"URL":"https:\/\/doi.org\/10.64898\/2026.08.01.742229","relation":{},"subject":[],"published":{"date-parts":[[2026,8,6]]},"subtype":"preprint"}}