{
  "id": 669488,
  "title": "New Dataset for Geograph 'Type' Tags",
  "url": "/competitions/predict-geographic-context-from-landscape-photos/discussion/669488",
  "author_name": "Barry Hunter",
  "post_date": "2026-01-22T16:33:45.638000",
  "votes": 1,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Inspired by the work on this competion have produced a new similar dataset: \n<a href=\"https://www.kaggle.com/datasets/barrybhunter/geograph-types-dataset-1/\" target=\"_blank\">https://www.kaggle.com/datasets/barrybhunter/geograph-types-dataset-1/</a></p>\n<p>Its again about predicting tags for landscape photos, but different set of tags. And instead of providing images for training, instead provides CLIP embeddings. With the idea can train a model using CLIP Embeddings as input, that offloads a lot of the visual understanding. </p>",
  "messages": [
    {
      "id": 3395287,
      "postDate": "2026-01-22T16:33:45.637Z",
      "content": "<p>Inspired by the work on this competion have produced a new similar dataset: \n<a href=\"https://www.kaggle.com/datasets/barrybhunter/geograph-types-dataset-1/\" target=\"_blank\">https://www.kaggle.com/datasets/barrybhunter/geograph-types-dataset-1/</a></p>\n<p>Its again about predicting tags for landscape photos, but different set of tags. And instead of providing images for training, instead provides CLIP embeddings. With the idea can train a model using CLIP Embeddings as input, that offloads a lot of the visual understanding. </p>",
      "rawMarkdown": "Inspired by the work on this competion have produced a new similar dataset: \nhttps://www.kaggle.com/datasets/barrybhunter/geograph-types-dataset-1/\n\nIts again about predicting tags for landscape photos, but different set of tags. And instead of providing images for training, instead provides CLIP embeddings. With the idea can train a model using CLIP Embeddings as input, that offloads a lot of the visual understanding. ",
      "votes": 1
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "3395287": "Inspired by the work on this competion have produced a new similar dataset: \nhttps://www.kaggle.com/datasets/barrybhunter/geograph-types-dataset-1/\n\nIts again about predicting tags for landscape photos, but different set of tags. And instead of providing images for training, instead provides CLIP embeddings. With the idea can train a model using CLIP Embeddings as input, that offloads a lot of the visual understanding. "
  }
}