{
  "id": 583347,
  "title": "Winning solutions code share / Non-medalist thoughts",
  "url": "/competitions/birdclef-2025/discussion/583347",
  "author_name": "Nikita Detkov",
  "post_date": "2025-06-06T06:58:03.305000",
  "votes": 6,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hi everyone! It's great to see how many people got fascinating results applying different techniques! Cudos to gold places! </p>\n<p>However, I was not particularly successful in this competition, just like the other 90% of us. I tried a lot of things, but I have never tried SED models, for example, and I never got to pseudo-labeling and knowledge distillation because I couldn't get past 0.8 with a single model. </p>\n<p>I see how the solutions and, what's more important, the code of previous years affect the discussion of this competition, and I encourage every top team to share! Not only inference notebooks, but also training code, pseudo-labeling, etc. — this will make us all better and more competitive.</p>\n<p>I know that a code of a person who is not in medals (me) doesn't count, but still. I share my GitHub repo with the code, experiments journal, results, and hypotheses — this is like a competition's diary, but it's reproducible. Here it is: <a href=\"https://github.com/detkov/BirdCLEFplus_2025\" target=\"_blank\">github.com/detkov/BirdCLEFplus_2025</a>. </p>\n<p>Hopefully, looking at my attempts, top teams will think \"weird flex but okay we can share something better than that\" 😀</p>",
  "messages": [
    {
      "id": 3218421,
      "postDate": "2025-06-06T06:58:03.307Z",
      "content": "<p>Hi everyone! It's great to see how many people got fascinating results applying different techniques! Cudos to gold places! </p>\n<p>However, I was not particularly successful in this competition, just like the other 90% of us. I tried a lot of things, but I have never tried SED models, for example, and I never got to pseudo-labeling and knowledge distillation because I couldn't get past 0.8 with a single model. </p>\n<p>I see how the solutions and, what's more important, the code of previous years affect the discussion of this competition, and I encourage every top team to share! Not only inference notebooks, but also training code, pseudo-labeling, etc. — this will make us all better and more competitive.</p>\n<p>I know that a code of a person who is not in medals (me) doesn't count, but still. I share my GitHub repo with the code, experiments journal, results, and hypotheses — this is like a competition's diary, but it's reproducible. Here it is: <a href=\"https://github.com/detkov/BirdCLEFplus_2025\" target=\"_blank\">github.com/detkov/BirdCLEFplus_2025</a>. </p>\n<p>Hopefully, looking at my attempts, top teams will think \"weird flex but okay we can share something better than that\" 😀</p>",
      "rawMarkdown": "Hi everyone! It's great to see how many people got fascinating results applying different techniques! Cudos to gold places! \n\nHowever, I was not particularly successful in this competition, just like the other 90% of us. I tried a lot of things, but I have never tried SED models, for example, and I never got to pseudo-labeling and knowledge distillation because I couldn't get past 0.8 with a single model. \n\nI see how the solutions and, what's more important, the code of previous years affect the discussion of this competition, and I encourage every top team to share! Not only inference notebooks, but also training code, pseudo-labeling, etc. — this will make us all better and more competitive.\n\nI know that a code of a person who is not in medals (me) doesn't count, but still. I share my GitHub repo with the code, experiments journal, results, and hypotheses — this is like a competition's diary, but it's reproducible. Here it is: [github.com/detkov/BirdCLEFplus_2025](https://github.com/detkov/BirdCLEFplus_2025). \n\nHopefully, looking at my attempts, top teams will think \"weird flex but okay we can share something better than that\" 😀\n",
      "votes": 6
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "3218421": "Hi everyone! It's great to see how many people got fascinating results applying different techniques! Cudos to gold places! \n\nHowever, I was not particularly successful in this competition, just like the other 90% of us. I tried a lot of things, but I have never tried SED models, for example, and I never got to pseudo-labeling and knowledge distillation because I couldn't get past 0.8 with a single model. \n\nI see how the solutions and, what's more important, the code of previous years affect the discussion of this competition, and I encourage every top team to share! Not only inference notebooks, but also training code, pseudo-labeling, etc. — this will make us all better and more competitive.\n\nI know that a code of a person who is not in medals (me) doesn't count, but still. I share my GitHub repo with the code, experiments journal, results, and hypotheses — this is like a competition's diary, but it's reproducible. Here it is: [github.com/detkov/BirdCLEFplus_2025](https://github.com/detkov/BirdCLEFplus_2025). \n\nHopefully, looking at my attempts, top teams will think \"weird flex but okay we can share something better than that\" 😀\n"
  }
}