{"paper":{"arxiv_id":"2201.03545","title":"A ConvNet for the 2020s","abstract":"The 'Roaring 20s' of visual recognition began with the introduction of Vision Transformers (ViTs), which quickly superseded ConvNets as the state-of-the-art image classification model. A vanilla ViT, on the other hand, faces difficulties when applied to general computer vision tasks such as object detection and semantic segmentation. In this work, we reexamine the design spaces and test the limits of what a pure ConvNet can achieve. We gradually 'modernize' a standard ResNet toward the design of a vision Transformer, and discover several key components that contribute to the performance difference along the way. The outcome is a family of pure ConvNet models dubbed ConvNeXt. Constructed entirely from standard ConvNet modules, ConvNeXts compete favorably with Transformers in terms of accuracy and scalability, achieving 87.8% ImageNet top-1 accuracy and outperforming Swin Transformers on COCO detection and ADE20K segmentation, while maintaining the simplicity and efficiency of standard ConvNets.","primary_category":"cs.CV","venue":"CVPR 2022","published_at":null,"latest_version":1,"withdrawn":false},"latest_version":{"id":"6ff412ac-6c7d-4882-8872-69d3ec774440","version":1,"source_url":"https://arxiv.org/abs/2201.03545","rendered_html_url":null,"rendering_engine":null},"verdict":{"id":"03e60e00-09b0-4782-8a76-c295166fcdfd","kind":"POST","status":"not_attempted","score":null,"confidence":null,"agent_version":"v0.1.0-convnext-imagenet-microslice","computed_at":"2026-05-14T23:24:31.282Z","is_current":true,"claim_citation":{"paper_arxiv_id":"2201.03545","section":"Table 1","row":"ConvNeXt-T","column":"ImageNet val top-1","reported_value":82.1,"reported_metric":"accuracy","quoted_text":"ConvNeXt-T 82.1","pdf_page":5,"notes":"Table 1 of arXiv:2201.03545 reports ConvNeXt-T ImageNet val top-1 = 82.1%. The matching HuggingFace checkpoint is `facebook/convnext-tiny-224`. Driver evaluates a 900-sample ImageNet val micro-slice. PROTOCOL_MATCH is `proxy`."},"protocol_match":"proxy"},"verdicts":{"post":{"id":"03e60e00-09b0-4782-8a76-c295166fcdfd","kind":"POST","status":"not_attempted","score":null,"confidence":null,"agent_version":"v0.1.0-convnext-imagenet-microslice","computed_at":"2026-05-14T23:24:31.282Z","is_current":true,"claim_citation":{"paper_arxiv_id":"2201.03545","section":"Table 1","row":"ConvNeXt-T","column":"ImageNet val top-1","reported_value":82.1,"reported_metric":"accuracy","quoted_text":"ConvNeXt-T 82.1","pdf_page":5,"notes":"Table 1 of arXiv:2201.03545 reports ConvNeXt-T ImageNet val top-1 = 82.1%. The matching HuggingFace checkpoint is `facebook/convnext-tiny-224`. Driver evaluates a 900-sample ImageNet val micro-slice. PROTOCOL_MATCH is `proxy`."},"protocol_match":"proxy"},"pre":null}}