{
  "id": 297429,
  "title": "Consused about yolox output bbox coordinates.",
  "url": "/competitions/tensorflow-great-barrier-reef/discussion/297429",
  "author_name": "",
  "post_date": "2021-12-27T09:25:56.718329200Z",
  "votes": 2,
  "comment_count": 4,
  "views": 0,
  "content": "<p>The highest lb notebook <a href=\"https://www.kaggle.com/parapapapam/yolox-inference-tracking-on-cots-lb-0-539\" target=\"_blank\">https://www.kaggle.com/parapapapam/yolox-inference-tracking-on-cots-lb-0-539</a> has one line of code at inference phase:</p>\n<pre><code>bboxes /= min(test_size[0] / img.shape[0], test_size[1] / img.shape[1])\n</code></pre>\n<p>The training script of the notebook use 960 x 960 size, so the scale for x and y are different.</p>\n<p>So why are the x and y are scaled back with the same factor? should them be different?</p>",
  "messages": [
    {
      "id": "1630416",
      "postDate": "12/27/2021 09:25:56",
      "content": "<p>The highest lb notebook <a href=\"https://www.kaggle.com/parapapapam/yolox-inference-tracking-on-cots-lb-0-539\" target=\"_blank\">https://www.kaggle.com/parapapapam/yolox-inference-tracking-on-cots-lb-0-539</a> has one line of code at inference phase:</p>\n<pre><code>bboxes /= min(test_size[0] / img.shape[0], test_size[1] / img.shape[1])\n</code></pre>\n<p>The training script of the notebook use 960 x 960 size, so the scale for x and y are different.</p>\n<p>So why are the x and y are scaled back with the same factor? should them be different?</p>",
      "rawMarkdown": "The highest lb notebook https://www.kaggle.com/parapapapam/yolox-inference-tracking-on-cots-lb-0-539 has one line of code at inference phase:\n\n```python\nbboxes /= min(test_size[0] / img.shape[0], test_size[1] / img.shape[1])\n```\n\nThe training script of the notebook use 960 x 960 size, so the scale for x and y are different.\n\nSo why are the x and y are scaled back with the same factor? should them be different?",
      "votes": null
    },
    {
      "id": "1630460",
      "postDate": "12/27/2021 10:42:58",
      "content": "<p>This is due to internal YoloX resizing. Look here: <a href=\"https://github.com/Megvii-BaseDetection/YOLOX/blob/main/yolox/data/data_augment.py#:~:text=r%20%3D%20min(input_size,astype(np.uint8\" target=\"_blank\">YoloX preproc function</a></p>",
      "rawMarkdown": "This is due to internal YoloX resizing. Look here: [YoloX preproc function](https://github.com/Megvii-BaseDetection/YOLOX/blob/main/yolox/data/data_augment.py#:~:text=r%20%3D%20min(input_size,astype(np.uint8)",
      "votes": null
    },
    {
      "id": "1630538",
      "postDate": "12/27/2021 12:29:23",
      "content": "<p>That's it! Thanks soooo much!</p>",
      "rawMarkdown": "That's it! Thanks soooo much!",
      "votes": null
    },
    {
      "id": "1630543",
      "postDate": "12/27/2021 12:31:14",
      "content": "<p>You are wlcome! Have a nice day and great competition ideas! 👍</p>",
      "rawMarkdown": "You are wlcome! Have a nice day and great competition ideas! 👍",
      "votes": null
    },
    {
      "id": "1630579",
      "postDate": "12/27/2021 13:13:29",
      "content": "<p>You just save my day bro, thank you again.</p>",
      "rawMarkdown": "You just save my day bro, thank you again.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1630460,
      "author_name": "remekkinas",
      "author_url": "",
      "post_date": "12/27/2021 10:42:58",
      "content": "<p>This is due to internal YoloX resizing. Look here: <a href=\"https://github.com/Megvii-BaseDetection/YOLOX/blob/main/yolox/data/data_augment.py#:~:text=r%20%3D%20min(input_size,astype(np.uint8\" target=\"_blank\">YoloX preproc function</a></p>",
      "votes": null,
      "replies": [
        {
          "id": 1630538,
          "author_name": "snaker",
          "author_url": "",
          "post_date": "12/27/2021 12:29:23",
          "content": "<p>That's it! Thanks soooo much!</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1630543,
          "author_name": "remekkinas",
          "author_url": "",
          "post_date": "12/27/2021 12:31:14",
          "content": "<p>You are wlcome! Have a nice day and great competition ideas! 👍</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1630579,
          "author_name": "snaker",
          "author_url": "",
          "post_date": "12/27/2021 13:13:29",
          "content": "<p>You just save my day bro, thank you again.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1630416": "The highest lb notebook https://www.kaggle.com/parapapapam/yolox-inference-tracking-on-cots-lb-0-539 has one line of code at inference phase:\n\n```python\nbboxes /= min(test_size[0] / img.shape[0], test_size[1] / img.shape[1])\n```\n\nThe training script of the notebook use 960 x 960 size, so the scale for x and y are different.\n\nSo why are the x and y are scaled back with the same factor? should them be different?",
    "1630460": "This is due to internal YoloX resizing. Look here: [YoloX preproc function](https://github.com/Megvii-BaseDetection/YOLOX/blob/main/yolox/data/data_augment.py#:~:text=r%20%3D%20min(input_size,astype(np.uint8)",
    "1630538": "That's it! Thanks soooo much!",
    "1630543": "You are wlcome! Have a nice day and great competition ideas! 👍",
    "1630579": "You just save my day bro, thank you again."
  },
  "source": "meta"
}