{
  "id": 359410,
  "title": "25th solution",
  "content": "<p>Thanks for organizers hosting a nice competition.</p>\n<p>Initially I was working on this competition with text-image contrastive method and trying dimension reduction technique like UMAP and also trying embedding each category(fashion/package/landmark…) on different discrete spaces, but these approach were not good.</p>\n<p>The approach of <a href=\"https://www.kaggle.com/code/motono0223/guie-clip-tensorflow-train-example\" target=\"_blank\">motono0223's baseline</a> was best for me. freezed CLIP + Arcface head.</p>\n<h2>Downsampling for the number of classes</h2>\n<ul>\n<li>64D embedding  space is not large so I tuned the number of classes for training.</li>\n<li>Training 4000 classes from Products10k/GLR was the best for me.<br>\n<img src=\"https://user-images.githubusercontent.com/61892693/195221946-f85f5e92-b57f-4315-bee9-4c9956d71a09.png\" alt=\"image\"></li>\n</ul>\n<h2>Hard to create good CV-LB correlation</h2>\n<ul>\n<li>I configured 6 different retrieval tasks, Products10k/GLR/Stanford Online Products/DeepFashion/MET/Food-101/ObjectNet but could not find clear CV-LB correlation.</li>\n<li>When only training with Products10k/GLR, Stanford Online Products(Cabinet/Sofa/Chair) retrieval setting was relatively correlated to LB. But it was still not perfect one.<br>\n<img src=\"https://user-images.githubusercontent.com/61892693/195222493-90ad0fef-b80a-46b2-bf9c-2e8f57028a3c.png\" alt=\"image\"></li>\n</ul>\n<h2>Freeze CLIP</h2>\n<ul>\n<li>Unfreeze CLIP finetune does not worked for me. Lowering learning rate got worked but freeze &amp; high learning rate was better.</li>\n<li>Fintuning other backbones, like Imagenet pretrained models, were not good.</li>\n<li>Adding more transformer layers on freezed CLIP was not good.</li>\n</ul>\n<h2>My code for this competition</h2>\n<ul>\n<li><a href=\"https://github.com/Fkaneko/kaggle_google_universal_image_embedding\" target=\"_blank\">https://github.com/Fkaneko/kaggle_google_universal_image_embedding</a></li>\n</ul>",
  "comments": [],
  "comment_count": 0,
  "messages": [
    {
      "id": 1983299,
      "postDate": "2022-10-12T00:37:54.667Z",
      "content": "<p>Thanks for organizers hosting a nice competition.</p>\n<p>Initially I was working on this competition with text-image contrastive method and trying dimension reduction technique like UMAP and also trying embedding each category(fashion/package/landmark…) on different discrete spaces, but these approach were not good.</p>\n<p>The approach of <a href=\"https://www.kaggle.com/code/motono0223/guie-clip-tensorflow-train-example\" target=\"_blank\">motono0223's baseline</a> was best for me. freezed CLIP + Arcface head.</p>\n<h2>Downsampling for the number of classes</h2>\n<ul>\n<li>64D embedding  space is not large so I tuned the number of classes for training.</li>\n<li>Training 4000 classes from Products10k/GLR was the best for me.<br>\n<img src=\"https://user-images.githubusercontent.com/61892693/195221946-f85f5e92-b57f-4315-bee9-4c9956d71a09.png\" alt=\"image\"></li>\n</ul>\n<h2>Hard to create good CV-LB correlation</h2>\n<ul>\n<li>I configured 6 different retrieval tasks, Products10k/GLR/Stanford Online Products/DeepFashion/MET/Food-101/ObjectNet but could not find clear CV-LB correlation.</li>\n<li>When only training with Products10k/GLR, Stanford Online Products(Cabinet/Sofa/Chair) retrieval setting was relatively correlated to LB. But it was still not perfect one.<br>\n<img src=\"https://user-images.githubusercontent.com/61892693/195222493-90ad0fef-b80a-46b2-bf9c-2e8f57028a3c.png\" alt=\"image\"></li>\n</ul>\n<h2>Freeze CLIP</h2>\n<ul>\n<li>Unfreeze CLIP finetune does not worked for me. Lowering learning rate got worked but freeze &amp; high learning rate was better.</li>\n<li>Fintuning other backbones, like Imagenet pretrained models, were not good.</li>\n<li>Adding more transformer layers on freezed CLIP was not good.</li>\n</ul>\n<h2>My code for this competition</h2>\n<ul>\n<li><a href=\"https://github.com/Fkaneko/kaggle_google_universal_image_embedding\" target=\"_blank\">https://github.com/Fkaneko/kaggle_google_universal_image_embedding</a></li>\n</ul>",
      "rawMarkdown": "Thanks for organizers hosting a nice competition.\n\nInitially I was working on this competition with text-image contrastive method and trying dimension reduction technique like UMAP and also trying embedding each category(fashion/package/landmark...) on different discrete spaces, but these approach were not good.\n\nThe approach of [motono0223's baseline](https://www.kaggle.com/code/motono0223/guie-clip-tensorflow-train-example) was best for me. freezed CLIP + Arcface head.\n\n## Downsampling for the number of classes\n* 64D embedding  space is not large so I tuned the number of classes for training.\n* Training 4000 classes from Products10k/GLR was the best for me.\n![image](https://user-images.githubusercontent.com/61892693/195221946-f85f5e92-b57f-4315-bee9-4c9956d71a09.png)\n\n## Hard to create good CV-LB correlation\n* I configured 6 different retrieval tasks, Products10k/GLR/Stanford Online Products/DeepFashion/MET/Food-101/ObjectNet but could not find clear CV-LB correlation.\n* When only training with Products10k/GLR, Stanford Online Products(Cabinet/Sofa/Chair) retrieval setting was relatively correlated to LB. But it was still not perfect one.\n![image](https://user-images.githubusercontent.com/61892693/195222493-90ad0fef-b80a-46b2-bf9c-2e8f57028a3c.png)\n## Freeze CLIP\n* Unfreeze CLIP finetune does not worked for me. Lowering learning rate got worked but freeze & high learning rate was better.\n* Fintuning other backbones, like Imagenet pretrained models, were not good.\n* Adding more transformer layers on freezed CLIP was not good.\n\n## My code for this competition\n* https://github.com/Fkaneko/kaggle_google_universal_image_embedding",
      "votes": 6
    }
  ],
  "complete": true,
  "source": "live",
  "fetched_at": "2026-10-22T08:32:35Z",
  "competition_id": 36414,
  "competition_slug": "google-universal-image-embedding",
  "team_id": 9039441,
  "solution_url": "/competitions/google-universal-image-embedding/writeups/fumihiro-kaneko-25th-solution"
}