{
  "id": 175845,
  "title": "Per-Image Error Analysis (1/3 of errors were missing annotations!)",
  "url": "/competitions/open-images-object-detection-rvc-2020/discussion/175845",
  "author_name": "Tyler Ganter",
  "post_date": "2020-08-19T15:23:27.448000",
  "votes": 1,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hey all,</p>\n<p>This seems like a very relevant place to mention this: I performed error analysis on Open Images with per-image evaluation (<em>using the same Tensorflow Object Detection API for this challenge</em>) and found a pretty crazy result that <strong>over 1/3 of errors were actually due to annotation error</strong>, not model error!</p>\n<p><a href=\"https://towardsdatascience.com/i-performed-error-analysis-on-open-images-and-now-i-have-trust-issues-89080e03ba09?source=friends_link&amp;sk=dec873d116b1bcc7fbd7d335640f1449\" target=\"_blank\">Here is the link</a> if you're interested!</p>\n<p>-Tyler</p>",
  "messages": [
    {
      "id": 977588,
      "postDate": "2020-08-19T15:23:27.450Z",
      "content": "<p>Hey all,</p>\n<p>This seems like a very relevant place to mention this: I performed error analysis on Open Images with per-image evaluation (<em>using the same Tensorflow Object Detection API for this challenge</em>) and found a pretty crazy result that <strong>over 1/3 of errors were actually due to annotation error</strong>, not model error!</p>\n<p><a href=\"https://towardsdatascience.com/i-performed-error-analysis-on-open-images-and-now-i-have-trust-issues-89080e03ba09?source=friends_link&amp;sk=dec873d116b1bcc7fbd7d335640f1449\" target=\"_blank\">Here is the link</a> if you're interested!</p>\n<p>-Tyler</p>",
      "rawMarkdown": "Hey all,\n\nThis seems like a very relevant place to mention this: I performed error analysis on Open Images with per-image evaluation (*using the same Tensorflow Object Detection API for this challenge*) and found a pretty crazy result that **over 1/3 of errors were actually due to annotation error**, not model error!\n\n[Here is the link](https://towardsdatascience.com/i-performed-error-analysis-on-open-images-and-now-i-have-trust-issues-89080e03ba09?source=friends_link&sk=dec873d116b1bcc7fbd7d335640f1449) if you're interested!\n\n-Tyler",
      "votes": 1
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "977588": "Hey all,\n\nThis seems like a very relevant place to mention this: I performed error analysis on Open Images with per-image evaluation (*using the same Tensorflow Object Detection API for this challenge*) and found a pretty crazy result that **over 1/3 of errors were actually due to annotation error**, not model error!\n\n[Here is the link](https://towardsdatascience.com/i-performed-error-analysis-on-open-images-and-now-i-have-trust-issues-89080e03ba09?source=friends_link&sk=dec873d116b1bcc7fbd7d335640f1449) if you're interested!\n\n-Tyler"
  }
}