{"paper":{"title":"YOLOX: Exceeding YOLO Series in 2021","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"YOLOX turns YOLO detectors anchor-free with a decoupled head and SimOTA assignment to reach higher accuracy at real-time speeds.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Feng Wang, Jian Sun, Songtao Liu, Zeming Li, Zheng Ge","submitted_at":"2021-07-18T12:55:11Z","abstract_excerpt":"In this report, we present some experienced improvements to YOLO series, forming a new high-performance detector -- YOLOX. We switch the YOLO detector to an anchor-free manner and conduct other advanced detection techniques, i.e., a decoupled head and the leading label assignment strategy SimOTA to achieve state-of-the-art results across a large scale range of models: For YOLO-Nano with only 0.91M parameters and 1.08G FLOPs, we get 25.3% AP on COCO, surpassing NanoDet by 1.8% AP; for YOLOv3, one of the most widely used detectors in industry, we boost it to 47.3% AP on COCO, outperforming the c"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"for YOLOX-L with roughly the same amount of parameters as YOLOv4-CSP, YOLOv5-L, we achieve 50.0% AP on COCO at a speed of 68.9 FPS on Tesla V100, exceeding YOLOv5-L by 1.8% AP.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the reported gains stem primarily from the anchor-free design, decoupled head, and SimOTA rather than from differences in training schedules, data augmentation, or hyperparameter choices versus the YOLOv5 and YOLOv4 baselines.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"YOLOX exceeds prior YOLO models by adopting anchor-free detection, decoupled heads, and SimOTA assignment to reach 50.0% AP on COCO for the large variant.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"YOLOX turns YOLO detectors anchor-free with a decoupled head and SimOTA assignment to reach higher accuracy at real-time speeds.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"23875a63c2afbdd499b8d891ea62ddd793d594628ffe4885a262a1f9e1dbdd6f"},"source":{"id":"2107.08430","kind":"arxiv","version":2},"verdict":{"id":"b98d8262-d0ae-4388-bb8b-7d4390baf489","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-13T10:27:58.064967Z","strongest_claim":"for YOLOX-L with roughly the same amount of parameters as YOLOv4-CSP, YOLOv5-L, we achieve 50.0% AP on COCO at a speed of 68.9 FPS on Tesla V100, exceeding YOLOv5-L by 1.8% AP.","one_line_summary":"YOLOX exceeds prior YOLO models by adopting anchor-free detection, decoupled heads, and SimOTA assignment to reach 50.0% AP on COCO for the large variant.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the reported gains stem primarily from the anchor-free design, decoupled head, and SimOTA rather than from differences in training schedules, data augmentation, or hyperparameter choices versus the YOLOv5 and YOLOv4 baselines.","pith_extraction_headline":"YOLOX turns YOLO detectors anchor-free with a decoupled head and SimOTA assignment to reach higher accuracy at real-time speeds."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2107.08430/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":40,"sample":[{"doi":"","year":2004,"title":"YOLOv4: Optimal Speed and Accuracy of Object Detection","work_id":"7057aaee-27f6-4209-a83c-f59727f937a8","ref_index":1,"cited_arxiv_id":"2004.10934","is_internal_anchor":true},{"doi":"","year":2020,"title":"End-to- end object detection with transformers","work_id":"554f512b-c3f2-430c-8d21-0cc7705ef850","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"You only look one-level feature","work_id":"4edbb038-a2ae-4c70-9133-3a04bb2282f1","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Ota: Optimal transport assignment for object detection","work_id":"bd871ef1-d3b6-4e0d-be6a-cc48768ac5bc","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Lla: Loss-aware label assignment for dense pedestrian detection","work_id":"b28e1fc5-2165-4c58-aca9-23ff38edc682","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":40,"snapshot_sha256":"3ed54aa125adb18e7fd94b7e036da8d89e584eed1be7e5cbf4549ebf0d233eba","internal_anchors":4},"formal_canon":{"evidence_count":2,"snapshot_sha256":"e9064f718d6c87deee7ddec859d99343e86d2c7f16064d178b750d9baefbd470"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}