{
  "id": 77765,
  "title": "Convergence of triplet loss",
  "url": "/competitions/humpback-whale-identification/discussion/77765",
  "author_name": "madonokoki",
  "post_date": "2019-01-16T14:09:34.903000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hi.</p>\n\n<p>I'm Implementing Siamese Network on pytorch and using triplet loss, I try to train with resnet (pretrained).\nBut it didn't converge yet. The loss is around the margin.</p>\n\n<p>I use data whose class number of Image is over  2 and exclude new whale.\nI try to use low lr (1e-6,1e-7) and <code>torch.nn.MarginRankingLoss</code> as a loss function. But it didn't be successed.</p>\n\n<p>If there are any problem, please tell me any idea.</p>\n\n<p>I set 0.2 as a margin of triplet loss</p>",
  "messages": [
    {
      "id": 456776,
      "postDate": "2019-01-16T14:09:34.903Z",
      "content": "<p>Hi.</p>\n\n<p>I'm Implementing Siamese Network on pytorch and using triplet loss, I try to train with resnet (pretrained).\nBut it didn't converge yet. The loss is around the margin.</p>\n\n<p>I use data whose class number of Image is over  2 and exclude new whale.\nI try to use low lr (1e-6,1e-7) and <code>torch.nn.MarginRankingLoss</code> as a loss function. But it didn't be successed.</p>\n\n<p>If there are any problem, please tell me any idea.</p>\n\n<p>I set 0.2 as a margin of triplet loss</p>",
      "rawMarkdown": "Hi.\n\nI'm Implementing Siamese Network on pytorch and using triplet loss, I try to train with resnet (pretrained).\nBut it didn't converge yet. The loss is around the margin.\n\nI use data whose class number of Image is over  2 and exclude new whale.\nI try to use low lr (1e-6,1e-7) and `torch.nn.MarginRankingLoss` as a loss function. But it didn't be successed.\n\nIf there are any problem, please tell me any idea.\n\nI set 0.2 as a margin of triplet loss"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "456776": "Hi.\n\nI'm Implementing Siamese Network on pytorch and using triplet loss, I try to train with resnet (pretrained).\nBut it didn't converge yet. The loss is around the margin.\n\nI use data whose class number of Image is over  2 and exclude new whale.\nI try to use low lr (1e-6,1e-7) and `torch.nn.MarginRankingLoss` as a loss function. But it didn't be successed.\n\nIf there are any problem, please tell me any idea.\n\nI set 0.2 as a margin of triplet loss"
  }
}