{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/imagenet-trained-cnns-are-biased-towards","title":"ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness","arxiv_id":"1811.12231","date":"2018-11-29","proceeding":"ICLR 2019 5","authors":["Robert Geirhos","Patricia Rubisch","Claudio Michaelis","Matthias Bethge","Felix A. Wichmann","Wieland Brendel"],"abstract":"Convolutional Neural Networks (CNNs) are commonly thought to recognise objects by learning increasingly complex representations of object shapes. Some recent studies suggest a more important role of image textures. We here put these conflicting hypotheses to a quantitative test by evaluating CNNs and human observers on images with a texture-shape cue conflict. We show that ImageNet-trained CNNs are strongly biased towards recognising textures rather than shapes, which is in stark contrast to human behavioural evidence and reveals fundamentally different classification strategies. We then demonstrate that the same standard architecture (ResNet-50) that learns a texture-based representation on ImageNet is able to learn a shape-based representation instead when trained on \"Stylized-ImageNet\", a stylized version of ImageNet. This provides a much better fit for human behavioural performance in our well-controlled psychophysical lab setting (nine experiments totalling 48,560 psychophysical trials across 97 observers) and comes with a number of unexpected emergent benefits such as improved object detection performance and previously unseen robustness towards a wide range of image distortions, highlighting advantages of a shape-based representation.","url_abs":"https://arxiv.org/abs/1811.12231v3","url_pdf":"https://arxiv.org/pdf/1811.12231v3.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"imagenet-trained-cnns-are-biased-towards","repo_url":"https://github.com/rgeirhos/Stylized-ImageNet","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"imagenet-trained-cnns-are-biased-towards","repo_url":"https://github.com/rgeirhos/texture-vs-shape","is_official":0,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok","spdx":"NOASSERTION"}},{"paper_slug":"imagenet-trained-cnns-are-biased-towards","repo_url":"https://github.com/LiYingwei/ShapeTextureDebiasedTraining","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"imagenet-trained-cnns-are-biased-towards","repo_url":"https://github.com/annstrange/breast-cancer-cnn","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok"}},{"paper_slug":"imagenet-trained-cnns-are-biased-towards","repo_url":"https://github.com/facebookresearch/augmentation-corruption","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"imagenet-trained-cnns-are-biased-towards","repo_url":"https://github.com/frank-roesler/Image_Segmentation","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}},{"paper_slug":"imagenet-trained-cnns-are-biased-towards","repo_url":"https://github.com/mbuet2ner/local-global-features-cnn","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"domain-generalization","task_name":"Domain Generalization"},{"task_slug":"image-classification","task_name":"Image Classification"},{"task_slug":"object-detection","task_name":"Object Detection"},{"task_slug":"object-recognition","task_name":"Object Recognition"},{"task_slug":"out-of-distribution-generalization","task_name":"Out-of-Distribution Generalization"},{"task_slug":"object-detection-1","task_name":"object-detection"}],"methods":[{"method_slug":"1x1-convolution","method_name":"1x1 Convolution"},{"method_slug":"average-pooling","method_name":"Average Pooling"},{"method_slug":"batch-normalization","method_name":"Batch Normalization"},{"method_slug":"bottleneck-residual-block","method_name":"Bottleneck Residual Block"},{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"global-average-pooling","method_name":"Global Average Pooling"},{"method_slug":"kaiming-initialization","method_name":"Kaiming Initialization"},{"method_slug":"max-pooling","method_name":"Max Pooling"},{"method_slug":"relu","method_name":"ReLU"},{"method_slug":"residual-block","method_name":"Residual Block"},{"method_slug":"residual-connection","method_name":"Residual Connection"}],"datasets_introduced":[{"slug":"stylized-imagenet","name":"Stylized ImageNet","full_name":"Stylized ImageNet"},{"slug":"shape-bias","name":"shape bias","full_name":""}],"methods_introduced":[],"results":[{"leaderboard":"/sota/domain-generalization-on-imagenet-a","task":"Domain Generalization","dataset":"ImageNet-A","model":"Stylized ImageNet (ResNet-50)","rank_in_archive_order":38,"of":39,"metrics":{"Top-1 accuracy %":"2.3"},"uses_additional_data":true},{"leaderboard":"/sota/domain-generalization-on-imagenet-c","task":"Domain Generalization","dataset":"ImageNet-C","model":"Stylized ImageNet (ResNet-50)","rank_in_archive_order":39,"of":47,"metrics":{"mean Corruption Error (mCE)":"69.3"},"uses_additional_data":true},{"leaderboard":"/sota/domain-generalization-on-imagenet-r","task":"Domain Generalization","dataset":"ImageNet-R","model":"Stylized ImageNet (ResNet-50)","rank_in_archive_order":34,"of":39,"metrics":{"Top-1 Error Rate":"58.5 "},"uses_additional_data":true},{"leaderboard":"/sota/domain-generalization-on-vizwiz","task":"Domain Generalization","dataset":"VizWiz-Classification","model":"ResNet-50 (SIN_IN_IN)","rank_in_archive_order":40,"of":90,"metrics":{"Accuracy - All Images":"39.2","Accuracy - Clean Images":"44.6","Accuracy - Corrupted Images":"32.4"},"uses_additional_data":false},{"leaderboard":"/sota/domain-generalization-on-vizwiz","task":"Domain Generalization","dataset":"VizWiz-Classification","model":"ResNet-50 (SIN_IN)","rank_in_archive_order":49,"of":90,"metrics":{"Accuracy - All Images":"38.2","Accuracy - Clean Images":"42.7","Accuracy - Corrupted Images":"32.5"},"uses_additional_data":false},{"leaderboard":"/sota/domain-generalization-on-vizwiz","task":"Domain Generalization","dataset":"VizWiz-Classification","model":"ResNet-50 (SIN)","rank_in_archive_order":86,"of":90,"metrics":{"Accuracy - All Images":"25.3","Accuracy - Clean Images":"30","Accuracy - Corrupted Images":"20.4"},"uses_additional_data":false},{"leaderboard":"/sota/object-recognition-on-shape-bias","task":"Object Recognition","dataset":"shape bias","model":"AlexNet","rank_in_archive_order":11,"of":18,"metrics":{"shape bias":"42.9"},"uses_additional_data":false},{"leaderboard":"/sota/object-recognition-on-shape-bias","task":"Object Recognition","dataset":"shape bias","model":"GoogLeNet","rank_in_archive_order":15,"of":18,"metrics":{"shape bias":"31.2"},"uses_additional_data":false},{"leaderboard":"/sota/object-recognition-on-shape-bias","task":"Object Recognition","dataset":"shape bias","model":"ResNet-50","rank_in_archive_order":17,"of":18,"metrics":{"shape bias":"22.1"},"uses_additional_data":false},{"leaderboard":"/sota/object-recognition-on-shape-bias","task":"Object Recognition","dataset":"shape bias","model":"VGG-16","rank_in_archive_order":18,"of":18,"metrics":{"shape bias":"17.2"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1811.12231","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"1811.12231"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-24T18:15:14+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"deterministic:regex_extraction","url":"https://github.com/rgeirhos/Stylized-ImageNet","reach":{"status":"ok","spdx":"MIT"}},{"provenance":"deterministic:regex_extraction","url":"https://github.com/rgeirhos/texture-vs-shape","reach":{"status":"ok","spdx":"NOASSERTION"}},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/facebookresearch/augmentation-corruption","reach":null},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/annstrange/breast-cancer-cnn","reach":{"status":"ok"}},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/mbuet2ner/local-global-features-cnn","reach":{"status":"ok","spdx":"MIT"}},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/LiYingwei/ShapeTextureDebiasedTraining","reach":null},{"provenance":"deterministic:regex_extraction","url":"https://github.com/BVLC/caffe","reach":{"status":"unanswered"}},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/frank-roesler/Image_Segmentation","reach":{"status":"ok"}},{"provenance":"deterministic:regex_extraction","url":"https://github.com/openimages/dataset","reach":{"status":"ok","spdx":"Apache-2.0"}}],"summary":{"unverified":6},"by_repo_kind":{"listed":{"samples":6,"ran":0,"repositories":2}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":0,"samples":[{"code_sha256_prefix":"c0af04d826190380","entry":"channels_first","repo":"mbuet2ner/local-global-features-cnn","repo_kind":"listed","path":"utils/ScrambleImage.py","file_url":"https://github.com/mbuet2ner/local-global-features-cnn/blob/HEAD/utils/ScrambleImage.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"c0af04d826190380"}},{"code_sha256_prefix":"fab2f87b04e9888a","entry":"channels_last","repo":"mbuet2ner/local-global-features-cnn","repo_kind":"listed","path":"utils/ScrambleImage.py","file_url":"https://github.com/mbuet2ner/local-global-features-cnn/blob/HEAD/utils/ScrambleImage.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"fab2f87b04e9888a"}},{"code_sha256_prefix":"586aecacc7ce053c","entry":"plot","repo":"mbuet2ner/local-global-features-cnn","repo_kind":"listed","path":"utils/utils.py","file_url":"https://github.com/mbuet2ner/local-global-features-cnn/blob/HEAD/utils/utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"586aecacc7ce053c"}},{"code_sha256_prefix":"ebcb545a49bf2ee2","entry":"resnext101","repo":"LiYingwei/ShapeTextureDebiasedTraining","repo_kind":"listed","path":"models/imagenet/resnext.py","file_url":"https://github.com/LiYingwei/ShapeTextureDebiasedTraining/blob/HEAD/models/imagenet/resnext.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":"MISDECLARED","metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"ebcb545a49bf2ee2"}},{"code_sha256_prefix":"fc281a9b16611bd2","entry":"resnext152","repo":"LiYingwei/ShapeTextureDebiasedTraining","repo_kind":"listed","path":"models/imagenet/resnext.py","file_url":"https://github.com/LiYingwei/ShapeTextureDebiasedTraining/blob/HEAD/models/imagenet/resnext.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"fc281a9b16611bd2"}},{"code_sha256_prefix":"2c55864d4a9f367c","entry":"resnext50","repo":"LiYingwei/ShapeTextureDebiasedTraining","repo_kind":"listed","path":"models/imagenet/resnext.py","file_url":"https://github.com/LiYingwei/ShapeTextureDebiasedTraining/blob/HEAD/models/imagenet/resnext.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"2c55864d4a9f367c"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}