{
  "id": 84777,
  "title": "New Paper on Arxiv that surveys some common ReID training tricks",
  "url": "/competitions/humpback-whale-identification/discussion/84777",
  "author_name": "",
  "post_date": "2019-03-19T15:57:40.745336Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I know the competition is already over. But I found a pretty well-written paper that helped explain the training tricks in a different way than some of the top write-ups. I found another perspective helpful in understanding. </p>\n\n<p><a href=\"https://arxiv.org/pdf/1903.07071.pdf\">https://arxiv.org/pdf/1903.07071.pdf</a></p>\n\n<p>Did you implement all of these tricks? What ones did they miss?\nI'd be interested in knowing how many people tried/used training their models with Random Erasing Augmentation. Did anyone do something similar to BNNeck and applied Batch Norm to their global feature vector for ID loss?</p>\n\n<p>-Paul</p>",
  "messages": [
    {
      "id": "494249",
      "postDate": "03/19/2019 15:57:40",
      "content": "<p>I know the competition is already over. But I found a pretty well-written paper that helped explain the training tricks in a different way than some of the top write-ups. I found another perspective helpful in understanding. </p>\n\n<p><a href=\"https://arxiv.org/pdf/1903.07071.pdf\">https://arxiv.org/pdf/1903.07071.pdf</a></p>\n\n<p>Did you implement all of these tricks? What ones did they miss?\nI'd be interested in knowing how many people tried/used training their models with Random Erasing Augmentation. Did anyone do something similar to BNNeck and applied Batch Norm to their global feature vector for ID loss?</p>\n\n<p>-Paul</p>",
      "rawMarkdown": "I know the competition is already over. But I found a pretty well-written paper that helped explain the training tricks in a different way than some of the top write-ups. I found another perspective helpful in understanding. \n\nhttps://arxiv.org/pdf/1903.07071.pdf\n\nDid you implement all of these tricks? What ones did they miss?\nI'd be interested in knowing how many people tried/used training their models with Random Erasing Augmentation. Did anyone do something similar to BNNeck and applied Batch Norm to their global feature vector for ID loss?\n\n-Paul",
      "votes": null
    },
    {
      "id": "495262",
      "postDate": "03/20/2019 22:06:39",
      "content": "<p>Just a quick glance, a lot of tricks in this paper and the 1st solution used ideas in this repo: <a href=\"https://github.com/L1aoXingyu/reid_baseline\">https://github.com/L1aoXingyu/reid_baseline</a></p>",
      "rawMarkdown": "Just a quick glance, a lot of tricks in this paper and the 1st solution used ideas in this repo: https://github.com/L1aoXingyu/reid_baseline",
      "votes": null
    },
    {
      "id": "500796",
      "postDate": "03/26/2019 15:07:27",
      "content": "<p>Thanks for your sharing! Reading it...</p>",
      "rawMarkdown": "Thanks for your sharing! Reading it...",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 495262,
      "author_name": "jihangz",
      "author_url": "",
      "post_date": "03/20/2019 22:06:39",
      "content": "<p>Just a quick glance, a lot of tricks in this paper and the 1st solution used ideas in this repo: <a href=\"https://github.com/L1aoXingyu/reid_baseline\">https://github.com/L1aoXingyu/reid_baseline</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 500796,
      "author_name": "benwu232",
      "author_url": "",
      "post_date": "03/26/2019 15:07:27",
      "content": "<p>Thanks for your sharing! Reading it...</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "494249": "I know the competition is already over. But I found a pretty well-written paper that helped explain the training tricks in a different way than some of the top write-ups. I found another perspective helpful in understanding. \n\nhttps://arxiv.org/pdf/1903.07071.pdf\n\nDid you implement all of these tricks? What ones did they miss?\nI'd be interested in knowing how many people tried/used training their models with Random Erasing Augmentation. Did anyone do something similar to BNNeck and applied Batch Norm to their global feature vector for ID loss?\n\n-Paul",
    "495262": "Just a quick glance, a lot of tricks in this paper and the 1st solution used ideas in this repo: https://github.com/L1aoXingyu/reid_baseline",
    "500796": "Thanks for your sharing! Reading it..."
  },
  "source": "meta"
}