{
  "id": 99989,
  "title": "How to improve speed? Yolo-v3",
  "url": "/competitions/open-images-2019-object-detection/discussion/99989",
  "author_name": "",
  "post_date": "2019-07-15T23:33:42.043300600Z",
  "votes": 11,
  "comment_count": 15,
  "views": 0,
  "content": "<p>Hi all,\nI have built a yolo-v3 running quite well on accuracy. However, it takes about 30s to process a picture. Isn't yolo a very fast model? Are there any techniques to improve speed?\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2368928%2F20c35e0abe3faaed605441e888857320%2FScreenshot%202019-07-16%20at%2000.19.23.png?generation=1563233492063291&amp;alt=media\" alt=\"\">\nAlso, due to the weight file that I used (<a href=\"https://pjreddie.com/media/files/yolov3.weights\">https://pjreddie.com/media/files/yolov3.weights</a>), only around 80 labels are available. Is there any way to add more labels?</p>\n\n<p>PS: Open to team forming, if you can help with my model, or I am handy to yours.</p>",
  "messages": [
    {
      "id": "576720",
      "postDate": "07/15/2019 23:33:42",
      "content": "<p>Hi all,\nI have built a yolo-v3 running quite well on accuracy. However, it takes about 30s to process a picture. Isn't yolo a very fast model? Are there any techniques to improve speed?\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2368928%2F20c35e0abe3faaed605441e888857320%2FScreenshot%202019-07-16%20at%2000.19.23.png?generation=1563233492063291&amp;alt=media\" alt=\"\">\nAlso, due to the weight file that I used (<a href=\"https://pjreddie.com/media/files/yolov3.weights\">https://pjreddie.com/media/files/yolov3.weights</a>), only around 80 labels are available. Is there any way to add more labels?</p>\n\n<p>PS: Open to team forming, if you can help with my model, or I am handy to yours.</p>",
      "rawMarkdown": "Hi all,\nI have built a yolo-v3 running quite well on accuracy. However, it takes about 30s to process a picture. Isn't yolo a very fast model? Are there any techniques to improve speed?\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2368928%2F20c35e0abe3faaed605441e888857320%2FScreenshot%202019-07-16%20at%2000.19.23.png?generation=1563233492063291&amp;alt=media)\nAlso, due to the weight file that I used (https://pjreddie.com/media/files/yolov3.weights), only around 80 labels are available. Is there any way to add more labels?\n\nPS: Open to team forming, if you can help with my model, or I am handy to yours.",
      "votes": null
    },
    {
      "id": "576749",
      "postDate": "07/16/2019 00:54:23",
      "content": "<p>are you using gpu? yolov3 is fast only on gpu. also, he made a pre-trained model for oid. look for it.... the model you are using is for coco</p>",
      "rawMarkdown": "are you using gpu? yolov3 is fast only on gpu. also, he made a pre-trained model for oid. look for it.... the model you are using is for coco",
      "votes": null
    },
    {
      "id": "577010",
      "postDate": "07/16/2019 08:40:52",
      "content": "<p>Who made a pre-trained model for oid?</p>",
      "rawMarkdown": "Who made a pre-trained model for oid?",
      "votes": null
    },
    {
      "id": "577680",
      "postDate": "07/17/2019 00:02:21",
      "content": "<p>Try searching for \"open images\" on the yolov3 home page? </p>",
      "rawMarkdown": "Try searching for \"open images\" on the yolov3 home page?",
      "votes": null
    },
    {
      "id": "578235",
      "postDate": "07/17/2019 13:54:51",
      "content": "<p>It's here : </p>\n\n<p><strong>YOLOv3 on the Open Images dataset</strong>\n```\nwget <a href=\"https://pjreddie.com/media/files/yolov3-openimages.weights\">https://pjreddie.com/media/files/yolov3-openimages.weights</a></p>\n\n<p>./darknet detector test cfg/openimages.data cfg/yolov3-openimages.cfg yolov3-openimages.weights\n```</p>",
      "rawMarkdown": "It's here : \n\n**YOLOv3 on the Open Images dataset**\n```\nwget https://pjreddie.com/media/files/yolov3-openimages.weights\n\n./darknet detector test cfg/openimages.data cfg/yolov3-openimages.cfg yolov3-openimages.weights\n```",
      "votes": null
    },
    {
      "id": "578278",
      "postDate": "07/17/2019 14:43:55",
      "content": "<p>thx</p>",
      "rawMarkdown": "thx",
      "votes": null
    },
    {
      "id": "578777",
      "postDate": "07/18/2019 06:28:44",
      "content": "<p>interesting</p>",
      "rawMarkdown": "interesting",
      "votes": null
    },
    {
      "id": "578832",
      "postDate": "07/18/2019 07:32:20",
      "content": "<p>cool</p>",
      "rawMarkdown": "cool",
      "votes": null
    },
    {
      "id": "579074",
      "postDate": "07/18/2019 13:14:29",
      "content": "<p>30s per image? I thought yolo-v3 is very fast.</p>",
      "rawMarkdown": "30s per image? I thought yolo-v3 is very fast.",
      "votes": null
    },
    {
      "id": "579199",
      "postDate": "07/18/2019 15:47:10",
      "content": "<p>It is 3s now after optimisation. \nHowever, still slow. </p>",
      "rawMarkdown": "It is 3s now after optimisation. \nHowever, still slow.",
      "votes": null
    },
    {
      "id": "579200",
      "postDate": "07/18/2019 15:47:23",
      "content": "<p>thx</p>",
      "rawMarkdown": "thx",
      "votes": null
    },
    {
      "id": "579201",
      "postDate": "07/18/2019 15:47:27",
      "content": "<p>thx</p>",
      "rawMarkdown": "thx",
      "votes": null
    },
    {
      "id": "579764",
      "postDate": "07/19/2019 07:19:20",
      "content": "<p>thx</p>",
      "rawMarkdown": "thx",
      "votes": null
    },
    {
      "id": "620719",
      "postDate": "09/07/2019 22:37:52",
      "content": "<p>Is the model any good for this new test dataset? <a href=\"/ratthachat\">@ratthachat</a> </p>",
      "rawMarkdown": "Is the model any good for this new test dataset? @ratthachat",
      "votes": null
    },
    {
      "id": "620740",
      "postDate": "09/07/2019 23:40:00",
      "content": "<p>Hi <a href=\"/overratedgman\">@overratedgman</a> Ovsen, I haven't yet participated in this comp. , so I do not know the answer :)</p>",
      "rawMarkdown": "Hi @overratedgman Ovsen, I haven't yet participated in this comp. , so I do not know the answer :)",
      "votes": null
    },
    {
      "id": "1159526",
      "postDate": "01/19/2021 10:31:55",
      "content": "<p>Hey. What kind of optimization did you use?</p>",
      "rawMarkdown": "Hey. What kind of optimization did you use?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 576749,
      "author_name": "moshel",
      "author_url": "",
      "post_date": "07/16/2019 00:54:23",
      "content": "<p>are you using gpu? yolov3 is fast only on gpu. also, he made a pre-trained model for oid. look for it.... the model you are using is for coco</p>",
      "votes": null,
      "replies": [
        {
          "id": 577010,
          "author_name": "yw6916",
          "author_url": "",
          "post_date": "07/16/2019 08:40:52",
          "content": "<p>Who made a pre-trained model for oid?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 577680,
          "author_name": "moshel",
          "author_url": "",
          "post_date": "07/17/2019 00:02:21",
          "content": "<p>Try searching for \"open images\" on the yolov3 home page? </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 578235,
          "author_name": "ratthachat",
          "author_url": "",
          "post_date": "07/17/2019 13:54:51",
          "content": "<p>It's here : </p>\n\n<p><strong>YOLOv3 on the Open Images dataset</strong>\n```\nwget <a href=\"https://pjreddie.com/media/files/yolov3-openimages.weights\">https://pjreddie.com/media/files/yolov3-openimages.weights</a></p>\n\n<p>./darknet detector test cfg/openimages.data cfg/yolov3-openimages.cfg yolov3-openimages.weights\n```</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 578278,
          "author_name": "yw6916",
          "author_url": "",
          "post_date": "07/17/2019 14:43:55",
          "content": "<p>thx</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 579764,
          "author_name": "yw6916",
          "author_url": "",
          "post_date": "07/19/2019 07:19:20",
          "content": "<p>thx</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 620719,
          "author_name": "overratedgman",
          "author_url": "",
          "post_date": "09/07/2019 22:37:52",
          "content": "<p>Is the model any good for this new test dataset? <a href=\"/ratthachat\">@ratthachat</a> </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 620740,
          "author_name": "ratthachat",
          "author_url": "",
          "post_date": "09/07/2019 23:40:00",
          "content": "<p>Hi <a href=\"/overratedgman\">@overratedgman</a> Ovsen, I haven't yet participated in this comp. , so I do not know the answer :)</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 578777,
      "author_name": "darbin",
      "author_url": "",
      "post_date": "07/18/2019 06:28:44",
      "content": "<p>interesting</p>",
      "votes": null,
      "replies": [
        {
          "id": 579201,
          "author_name": "yw6916",
          "author_url": "",
          "post_date": "07/18/2019 15:47:27",
          "content": "<p>thx</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 578832,
      "author_name": "fillerink",
      "author_url": "",
      "post_date": "07/18/2019 07:32:20",
      "content": "<p>cool</p>",
      "votes": null,
      "replies": [
        {
          "id": 579200,
          "author_name": "yw6916",
          "author_url": "",
          "post_date": "07/18/2019 15:47:23",
          "content": "<p>thx</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 579074,
      "author_name": "lasnalo",
      "author_url": "",
      "post_date": "07/18/2019 13:14:29",
      "content": "<p>30s per image? I thought yolo-v3 is very fast.</p>",
      "votes": null,
      "replies": [
        {
          "id": 579199,
          "author_name": "yw6916",
          "author_url": "",
          "post_date": "07/18/2019 15:47:10",
          "content": "<p>It is 3s now after optimisation. \nHowever, still slow. </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1159526,
          "author_name": "omar16100",
          "author_url": "",
          "post_date": "01/19/2021 10:31:55",
          "content": "<p>Hey. What kind of optimization did you use?</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "576720": "Hi all,\nI have built a yolo-v3 running quite well on accuracy. However, it takes about 30s to process a picture. Isn't yolo a very fast model? Are there any techniques to improve speed?\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2368928%2F20c35e0abe3faaed605441e888857320%2FScreenshot%202019-07-16%20at%2000.19.23.png?generation=1563233492063291&amp;alt=media)\nAlso, due to the weight file that I used (https://pjreddie.com/media/files/yolov3.weights), only around 80 labels are available. Is there any way to add more labels?\n\nPS: Open to team forming, if you can help with my model, or I am handy to yours.",
    "576749": "are you using gpu? yolov3 is fast only on gpu. also, he made a pre-trained model for oid. look for it.... the model you are using is for coco",
    "577010": "Who made a pre-trained model for oid?",
    "577680": "Try searching for \"open images\" on the yolov3 home page?",
    "578235": "It's here : \n\n**YOLOv3 on the Open Images dataset**\n```\nwget https://pjreddie.com/media/files/yolov3-openimages.weights\n\n./darknet detector test cfg/openimages.data cfg/yolov3-openimages.cfg yolov3-openimages.weights\n```",
    "578278": "thx",
    "578777": "interesting",
    "578832": "cool",
    "579074": "30s per image? I thought yolo-v3 is very fast.",
    "579199": "It is 3s now after optimisation. \nHowever, still slow.",
    "579200": "thx",
    "579201": "thx",
    "579764": "thx",
    "620719": "Is the model any good for this new test dataset? @ratthachat",
    "620740": "Hi @overratedgman Ovsen, I haven't yet participated in this comp. , so I do not know the answer :)",
    "1159526": "Hey. What kind of optimization did you use?"
  },
  "source": "meta"
}