{
  "id": 113001,
  "title": "Prediction visualisation",
  "url": "/competitions/3d-object-detection-for-autonomous-vehicles/discussion/113001",
  "author_name": "",
  "post_date": "2019-10-16T13:54:43.760906300Z",
  "votes": 2,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Does anyone have a straightforward way of visualizing predictions, i.e. overlaying predicted 3D boxes onto test images? It seems possible in the lyftdataset commands but I can't get it working with the model output!</p>",
  "messages": [
    {
      "id": "650550",
      "postDate": "10/16/2019 13:54:43",
      "content": "<p>Does anyone have a straightforward way of visualizing predictions, i.e. overlaying predicted 3D boxes onto test images? It seems possible in the lyftdataset commands but I can't get it working with the model output!</p>",
      "rawMarkdown": "Does anyone have a straightforward way of visualizing predictions, i.e. overlaying predicted 3D boxes onto test images? It seems possible in the lyftdataset commands but I can't get it working with the model output!",
      "votes": null
    },
    {
      "id": "650572",
      "postDate": "10/16/2019 14:13:39",
      "content": "<p>I've written some code for visualizing predictions but it's kinda messy at the moment, I'll clean it up and share with you, maybe as a kernel.  Right now I convert my raw predictions to <a href=\"https://github.com/lyft/nuscenes-devkit/blob/master/lyft_dataset_sdk/utils/data_classes.py#L478\">Boxes</a> in global frame and then do the visualization, you fine with this? Let me know if you have any ideas.</p>",
      "rawMarkdown": "I've written some code for visualizing predictions but it's kinda messy at the moment, I'll clean it up and share with you, maybe as a kernel.  Right now I convert my raw predictions to [Boxes](https://github.com/lyft/nuscenes-devkit/blob/master/lyft_dataset_sdk/utils/data_classes.py#L478) in global frame and then do the visualization, you fine with this? Let me know if you have any ideas.",
      "votes": null
    },
    {
      "id": "650644",
      "postDate": "10/16/2019 15:06:04",
      "content": "<p>Congratulations on your performance so far! That sounds like a good approach, it would be very helpful if you could please share this kernel :)</p>",
      "rawMarkdown": "Congratulations on your performance so far! That sounds like a good approach, it would be very helpful if you could please share this kernel :)",
      "votes": null
    },
    {
      "id": "650745",
      "postDate": "10/16/2019 16:40:48",
      "content": "<p>Here it is: <a href=\"https://www.kaggle.com/rishabhiitbhu/visualizing-predictions\">https://www.kaggle.com/rishabhiitbhu/visualizing-predictions</a></p>",
      "rawMarkdown": "Here it is: https://www.kaggle.com/rishabhiitbhu/visualizing-predictions",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 650572,
      "author_name": "rishabhiitbhu",
      "author_url": "",
      "post_date": "10/16/2019 14:13:39",
      "content": "<p>I've written some code for visualizing predictions but it's kinda messy at the moment, I'll clean it up and share with you, maybe as a kernel.  Right now I convert my raw predictions to <a href=\"https://github.com/lyft/nuscenes-devkit/blob/master/lyft_dataset_sdk/utils/data_classes.py#L478\">Boxes</a> in global frame and then do the visualization, you fine with this? Let me know if you have any ideas.</p>",
      "votes": null,
      "replies": [
        {
          "id": 650644,
          "author_name": "jake126",
          "author_url": "",
          "post_date": "10/16/2019 15:06:04",
          "content": "<p>Congratulations on your performance so far! That sounds like a good approach, it would be very helpful if you could please share this kernel :)</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 650745,
          "author_name": "rishabhiitbhu",
          "author_url": "",
          "post_date": "10/16/2019 16:40:48",
          "content": "<p>Here it is: <a href=\"https://www.kaggle.com/rishabhiitbhu/visualizing-predictions\">https://www.kaggle.com/rishabhiitbhu/visualizing-predictions</a></p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "650550": "Does anyone have a straightforward way of visualizing predictions, i.e. overlaying predicted 3D boxes onto test images? It seems possible in the lyftdataset commands but I can't get it working with the model output!",
    "650572": "I've written some code for visualizing predictions but it's kinda messy at the moment, I'll clean it up and share with you, maybe as a kernel.  Right now I convert my raw predictions to [Boxes](https://github.com/lyft/nuscenes-devkit/blob/master/lyft_dataset_sdk/utils/data_classes.py#L478) in global frame and then do the visualization, you fine with this? Let me know if you have any ideas.",
    "650644": "Congratulations on your performance so far! That sounds like a good approach, it would be very helpful if you could please share this kernel :)",
    "650745": "Here it is: https://www.kaggle.com/rishabhiitbhu/visualizing-predictions"
  },
  "source": "meta"
}