{
  "id": 254201,
  "title": "YOLOX: Exceeding YOLO Series in 2021",
  "url": "/competitions/siim-covid19-detection/discussion/254201",
  "author_name": "",
  "post_date": "2021-07-20T15:27:30.309691400Z",
  "votes": 9,
  "comment_count": 2,
  "views": 0,
  "content": "<p><img src=\"https://github.com/Megvii-BaseDetection/YOLOX/blob/main/assets/git_fig.png?raw=true\" alt=\"\"></p>\n<pre><code>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\nassignment strategy SimOTA to achieve state-of-the-art results across a large scale range of models: For YOLONano 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 current best practice by 3.0% AP; 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. Further, we won the 1st Place on Streaming Perception Challenge (Workshop on Autonomous Driving at CVPR 2021) using a single YOLOX-L model. We hope this report can provide useful experience for developers and researchers in practical scenes, and we also provide deploy versions with ONNX, TensorRT, NCNN, and Openvino supported\n</code></pre>\n<p>Code: <a href=\"https://github.com/Megvii-BaseDetection/YOLOX\" target=\"_blank\">https://github.com/Megvii-BaseDetection/YOLOX</a><br>\nPaper: <a href=\"https://arxiv.org/pdf/2107.08430v1.pdf\" target=\"_blank\">https://arxiv.org/pdf/2107.08430v1.pdf</a></p>",
  "messages": [
    {
      "id": "1394763",
      "postDate": "07/20/2021 15:27:30",
      "content": "<p><img src=\"https://github.com/Megvii-BaseDetection/YOLOX/blob/main/assets/git_fig.png?raw=true\" alt=\"\"></p>\n<pre><code>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\nassignment strategy SimOTA to achieve state-of-the-art results across a large scale range of models: For YOLONano 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 current best practice by 3.0% AP; 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. Further, we won the 1st Place on Streaming Perception Challenge (Workshop on Autonomous Driving at CVPR 2021) using a single YOLOX-L model. We hope this report can provide useful experience for developers and researchers in practical scenes, and we also provide deploy versions with ONNX, TensorRT, NCNN, and Openvino supported\n</code></pre>\n<p>Code: <a href=\"https://github.com/Megvii-BaseDetection/YOLOX\" target=\"_blank\">https://github.com/Megvii-BaseDetection/YOLOX</a><br>\nPaper: <a href=\"https://arxiv.org/pdf/2107.08430v1.pdf\" target=\"_blank\">https://arxiv.org/pdf/2107.08430v1.pdf</a></p>",
      "rawMarkdown": "![](https://github.com/Megvii-BaseDetection/YOLOX/blob/main/assets/git_fig.png?raw=true)\n```\nIn 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\nassignment strategy SimOTA to achieve state-of-the-art results across a large scale range of models: For YOLONano 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 current best practice by 3.0% AP; 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. Further, we won the 1st Place on Streaming Perception Challenge (Workshop on Autonomous Driving at CVPR 2021) using a single YOLOX-L model. We hope this report can provide useful experience for developers and researchers in practical scenes, and we also provide deploy versions with ONNX, TensorRT, NCNN, and Openvino supported\n```\nCode: https://github.com/Megvii-BaseDetection/YOLOX\nPaper: https://arxiv.org/pdf/2107.08430v1.pdf",
      "votes": null
    },
    {
      "id": "1399500",
      "postDate": "07/25/2021 11:30:22",
      "content": "<p>I didnt know that. thank you shareing.😄</p>",
      "rawMarkdown": "I didnt know that. thank you shareing.😄",
      "votes": null
    },
    {
      "id": "1400086",
      "postDate": "07/26/2021 03:23:02",
      "content": "<p>a method sewing up many techniques.<br>\nI expect someone to check it out in this competition.</p>",
      "rawMarkdown": "a method sewing up many techniques.\nI expect someone to check it out in this competition.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1399500,
      "author_name": "tensorchoko",
      "author_url": "",
      "post_date": "07/25/2021 11:30:22",
      "content": "<p>I didnt know that. thank you shareing.😄</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1400086,
      "author_name": "tiandaye",
      "author_url": "",
      "post_date": "07/26/2021 03:23:02",
      "content": "<p>a method sewing up many techniques.<br>\nI expect someone to check it out in this competition.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1394763": "![](https://github.com/Megvii-BaseDetection/YOLOX/blob/main/assets/git_fig.png?raw=true)\n```\nIn 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\nassignment strategy SimOTA to achieve state-of-the-art results across a large scale range of models: For YOLONano 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 current best practice by 3.0% AP; 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. Further, we won the 1st Place on Streaming Perception Challenge (Workshop on Autonomous Driving at CVPR 2021) using a single YOLOX-L model. We hope this report can provide useful experience for developers and researchers in practical scenes, and we also provide deploy versions with ONNX, TensorRT, NCNN, and Openvino supported\n```\nCode: https://github.com/Megvii-BaseDetection/YOLOX\nPaper: https://arxiv.org/pdf/2107.08430v1.pdf",
    "1399500": "I didnt know that. thank you shareing.😄",
    "1400086": "a method sewing up many techniques.\nI expect someone to check it out in this competition."
  },
  "source": "meta"
}