I wonder if the baseline results are based on the descriptors extracted from the pretrained Delf model. Can you please give some information about the baseline model (scoring 0.271)?
", "messages": [ { "id": "918917", "postDate": "07/07/2020 15:33:21", "content": "I wonder if the baseline results are based on the descriptors extracted from the pretrained Delf model. Can you please give some information about the baseline model (scoring 0.271)?
", "rawMarkdown": "I wonder if the baseline results are based on the descriptors extracted from the pretrained Delf model. Can you please give some information about the baseline model (scoring 0.271)?", "votes": null }, { "id": "919092", "postDate": "07/07/2020 17:47:04", "content": "Yes, you can see more information about the baseline model in its own page: https://www.kaggle.com/camaskew/baseline-landmark-retrieval-model
\n\nAs mentioned in the second submission post, this baseline model is using a different codebase for training (although generally similar).
", "rawMarkdown": "Yes, you can see more information about the baseline model in its own page: https://www.kaggle.com/camaskew/baseline-landmark-retrieval-model\n\nAs mentioned in the [second submission post](https://www.kaggle.com/c/landmark-retrieval-2020/discussion/163350), this baseline model is using a different codebase for training (although generally similar).", "votes": null }, { "id": "919704", "postDate": "07/08/2020 04:25:33", "content": "its very very hard to beat the baseline..
", "rawMarkdown": "its very very hard to beat the baseline..", "votes": null }, { "id": "923663", "postDate": "07/11/2020 04:20:43", "content": "@andrefaraujo \non DELF training repo, is there any option to use pre-trained weights? if we train on DELF, will the global descriptors included in it? or only local will be trained?
", "rawMarkdown": "andrefaraujo \non DELF training repo, is there any option to use pre-trained weights? if we train on DELF, will the global descriptors included in it? or only local will be trained?", "votes": null }, { "id": "927950", "postDate": "07/13/2020 16:49:47", "content": "You can see an example of loading pre-trained weights in the DELF training here.
\n\nThe current DELF training pipeline uses two heads, one for global feature learning and the other for local feature learning. So, there are global descriptors in it, yes, and the model can be exported for Kaggle format following these instructions.
", "rawMarkdown": "You can see an example of loading pre-trained weights in the DELF training [here](https://github.com/tensorflow/models/tree/master/research/delf/delf/python/training#running-the-training).\n\nThe current DELF training pipeline uses two heads, one for global feature learning and the other for local feature learning. So, there are global descriptors in it, yes, and the model can be exported for Kaggle format following [these instructions](https://github.com/tensorflow/models/tree/master/research/delf/delf/python/training#kaggle-compatible-global-feature-model).", "votes": null } ], "comments": [ { "id": 919092, "author_name": "andrefaraujo", "author_url": "", "post_date": "07/07/2020 17:47:04", "content": "Yes, you can see more information about the baseline model in its own page: https://www.kaggle.com/camaskew/baseline-landmark-retrieval-model
\n\nAs mentioned in the second submission post, this baseline model is using a different codebase for training (although generally similar).
", "votes": null, "replies": [] }, { "id": 919704, "author_name": "udaygurugubelli", "author_url": "", "post_date": "07/08/2020 04:25:33", "content": "its very very hard to beat the baseline..
", "votes": null, "replies": [] }, { "id": 923663, "author_name": "udaygurugubelli", "author_url": "", "post_date": "07/11/2020 04:20:43", "content": "@andrefaraujo \non DELF training repo, is there any option to use pre-trained weights? if we train on DELF, will the global descriptors included in it? or only local will be trained?
", "votes": null, "replies": [ { "id": 927950, "author_name": "andrefaraujo", "author_url": "", "post_date": "07/13/2020 16:49:47", "content": "You can see an example of loading pre-trained weights in the DELF training here.
\n\nThe current DELF training pipeline uses two heads, one for global feature learning and the other for local feature learning. So, there are global descriptors in it, yes, and the model can be exported for Kaggle format following these instructions.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "918917": "I wonder if the baseline results are based on the descriptors extracted from the pretrained Delf model. Can you please give some information about the baseline model (scoring 0.271)?", "919092": "Yes, you can see more information about the baseline model in its own page: https://www.kaggle.com/camaskew/baseline-landmark-retrieval-model\n\nAs mentioned in the [second submission post](https://www.kaggle.com/c/landmark-retrieval-2020/discussion/163350), this baseline model is using a different codebase for training (although generally similar).", "919704": "its very very hard to beat the baseline..", "923663": "andrefaraujo \non DELF training repo, is there any option to use pre-trained weights? if we train on DELF, will the global descriptors included in it? or only local will be trained?", "927950": "You can see an example of loading pre-trained weights in the DELF training [here](https://github.com/tensorflow/models/tree/master/research/delf/delf/python/training#running-the-training).\n\nThe current DELF training pipeline uses two heads, one for global feature learning and the other for local feature learning. So, there are global descriptors in it, yes, and the model can be exported for Kaggle format following [these instructions](https://github.com/tensorflow/models/tree/master/research/delf/delf/python/training#kaggle-compatible-global-feature-model)." }, "source": "meta" }