{
  "id": 119478,
  "title": "about 6d-vnet",
  "url": "/competitions/pku-autonomous-driving/discussion/119478",
  "author_name": "",
  "post_date": "2019-11-28T22:13:50.634180700Z",
  "votes": 7,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Does anybody succeeded training/predicting 6dof using\ncode <a href=\"https://github.com/stevenwudi/6DVNET\">https://github.com/stevenwudi/6DVNET</a>?</p>\n\n<p>After training for 30k iterations using config <code>configs/e2e_3d_car_101_FPN_triple_head.yaml</code>\nI got bad predictions of translation/rotation:</p>\n\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1095310%2F90dffa6dd8a6667a27430d87a04f6159%2Fkaggle1.png?generation=1574978752209002&amp;alt=media\" alt=\"\"></p>\n\n<p>I got also bad mask predictions but after debugging found a bug there with bad indexing of gtruth. after fixing bug got good mask prediction as in maskrcnn:</p>\n\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1095310%2F5e37eafec2f5b8342f899b9a81f8c6a8%2Fkaggle2.png?generation=1574979135959451&amp;alt=media\" alt=\"\"></p>\n\n<p>Tried also varying learning rate schedule but no much help.</p>",
  "messages": [
    {
      "id": "683843",
      "postDate": "11/28/2019 22:13:50",
      "content": "<p>Does anybody succeeded training/predicting 6dof using\ncode <a href=\"https://github.com/stevenwudi/6DVNET\">https://github.com/stevenwudi/6DVNET</a>?</p>\n\n<p>After training for 30k iterations using config <code>configs/e2e_3d_car_101_FPN_triple_head.yaml</code>\nI got bad predictions of translation/rotation:</p>\n\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1095310%2F90dffa6dd8a6667a27430d87a04f6159%2Fkaggle1.png?generation=1574978752209002&amp;alt=media\" alt=\"\"></p>\n\n<p>I got also bad mask predictions but after debugging found a bug there with bad indexing of gtruth. after fixing bug got good mask prediction as in maskrcnn:</p>\n\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1095310%2F5e37eafec2f5b8342f899b9a81f8c6a8%2Fkaggle2.png?generation=1574979135959451&amp;alt=media\" alt=\"\"></p>\n\n<p>Tried also varying learning rate schedule but no much help.</p>",
      "rawMarkdown": "Does anybody succeeded training/predicting 6dof using\ncode https://github.com/stevenwudi/6DVNET?\n\nAfter training for 30k iterations using config `configs/e2e_3d_car_101_FPN_triple_head.yaml`\nI got bad predictions of translation/rotation:\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1095310%2F90dffa6dd8a6667a27430d87a04f6159%2Fkaggle1.png?generation=1574978752209002&amp;alt=media)\n\nI got also bad mask predictions but after debugging found a bug there with bad indexing of gtruth. after fixing bug got good mask prediction as in maskrcnn:\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1095310%2F5e37eafec2f5b8342f899b9a81f8c6a8%2Fkaggle2.png?generation=1574979135959451&amp;alt=media)\n\nTried also varying learning rate schedule but no much help.",
      "votes": null
    },
    {
      "id": "684284",
      "postDate": "11/29/2019 14:31:58",
      "content": "<p>wow!\nvery impressive\nthanks for sharing this valuable post with us</p>",
      "rawMarkdown": "wow!\nvery impressive\nthanks for sharing this valuable post with us",
      "votes": null
    },
    {
      "id": "698600",
      "postDate": "12/19/2019 13:02:27",
      "content": "<p>can you tell me from where I can start and what strategy should I follow \nwhat I mean is that should I train the yolo or resnet or mask rcnn  which one first</p>",
      "rawMarkdown": "can you tell me from where I can start and what strategy should I follow \nwhat I mean is that should I train the yolo or resnet or mask rcnn  which one first",
      "votes": null
    },
    {
      "id": "698603",
      "postDate": "12/19/2019 13:04:04",
      "content": "<p>what I mean is that should I train Yolo first or resnet or should I train mask rcnn which one could you can guide me the strategy</p>",
      "rawMarkdown": "what I mean is that should I train Yolo first or resnet or should I train mask rcnn which one could you can guide me the strategy",
      "votes": null
    },
    {
      "id": "700601",
      "postDate": "12/22/2019 09:26:29",
      "content": "<p>Thanks for sharing this information.. </p>",
      "rawMarkdown": "Thanks for sharing this information..",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 684284,
      "author_name": "mobassir",
      "author_url": "",
      "post_date": "11/29/2019 14:31:58",
      "content": "<p>wow!\nvery impressive\nthanks for sharing this valuable post with us</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 698600,
      "author_name": "iftikharhumayun",
      "author_url": "",
      "post_date": "12/19/2019 13:02:27",
      "content": "<p>can you tell me from where I can start and what strategy should I follow \nwhat I mean is that should I train the yolo or resnet or mask rcnn  which one first</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 698603,
      "author_name": "iftikharhumayun",
      "author_url": "",
      "post_date": "12/19/2019 13:04:04",
      "content": "<p>what I mean is that should I train Yolo first or resnet or should I train mask rcnn which one could you can guide me the strategy</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 700601,
      "author_name": "swathym",
      "author_url": "",
      "post_date": "12/22/2019 09:26:29",
      "content": "<p>Thanks for sharing this information.. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "683843": "Does anybody succeeded training/predicting 6dof using\ncode https://github.com/stevenwudi/6DVNET?\n\nAfter training for 30k iterations using config `configs/e2e_3d_car_101_FPN_triple_head.yaml`\nI got bad predictions of translation/rotation:\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1095310%2F90dffa6dd8a6667a27430d87a04f6159%2Fkaggle1.png?generation=1574978752209002&amp;alt=media)\n\nI got also bad mask predictions but after debugging found a bug there with bad indexing of gtruth. after fixing bug got good mask prediction as in maskrcnn:\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1095310%2F5e37eafec2f5b8342f899b9a81f8c6a8%2Fkaggle2.png?generation=1574979135959451&amp;alt=media)\n\nTried also varying learning rate schedule but no much help.",
    "684284": "wow!\nvery impressive\nthanks for sharing this valuable post with us",
    "698600": "can you tell me from where I can start and what strategy should I follow \nwhat I mean is that should I train the yolo or resnet or mask rcnn  which one first",
    "698603": "what I mean is that should I train Yolo first or resnet or should I train mask rcnn which one could you can guide me the strategy",
    "700601": "Thanks for sharing this information.."
  },
  "source": "meta"
}