{ "id": 290992, "title": "EfficientDet", "url": "/competitions/tensorflow-great-barrier-reef/discussion/290992", "author_name": "", "post_date": "2021-11-27T07:16:28.327580800Z", "votes": 6, "comment_count": 4, "views": 0, "content": "

As always, I am not familier with the object detection model, so I surveyed briefly.
\nThen I found EficientDet.

\n

This model seems good because in the last competition called \"Global Wheat Detection\", grand master Mr @tereka won 5th place with this model.

\n

His discussion is available here.

\n

Also, the notebook for EfficientDet from @shonenkov is available here.

\n

So I prepared the training notebook.
\nThis is elementary notebook(version 0), so from now on I will implement some methods(e.g. other augmentation technique, Pseudo-Labeling and other if any)

\n

You cannot use the shonenkov's notebook for this competion straightforward.
\nSome modification are needed( e.g. dependencies such as albumentation and efficientdet libraries for me.)

\n

I am also preparing inference notebook in this weekend.
\nI will release it soon.

\n

P.S.
\nI made a mistake..
\nI created the Notebook with pytorch.
\nSo I will create tensorflow version.

", "messages": [ { "id": "1597064", "postDate": "11/27/2021 07:16:28", "content": "

As always, I am not familier with the object detection model, so I surveyed briefly.
\nThen I found EficientDet.

\n

This model seems good because in the last competition called \"Global Wheat Detection\", grand master Mr @tereka won 5th place with this model.

\n

His discussion is available here.

\n

Also, the notebook for EfficientDet from @shonenkov is available here.

\n

So I prepared the training notebook.
\nThis is elementary notebook(version 0), so from now on I will implement some methods(e.g. other augmentation technique, Pseudo-Labeling and other if any)

\n

You cannot use the shonenkov's notebook for this competion straightforward.
\nSome modification are needed( e.g. dependencies such as albumentation and efficientdet libraries for me.)

\n

I am also preparing inference notebook in this weekend.
\nI will release it soon.

\n

P.S.
\nI made a mistake..
\nI created the Notebook with pytorch.
\nSo I will create tensorflow version.

", "rawMarkdown": "As always, I am not familier with the object detection model, so I surveyed briefly.\nThen I found [**EficientDet**](https://arxiv.org/abs/1911.09070).\n\nThis model seems good because in the last competition called \"[Global Wheat Detection](https://www.kaggle.com/c/global-wheat-detection)\", grand master Mr @tereka won 5th place with this model.\n\nHis discussion is available [here](https://www.kaggle.com/c/global-wheat-detection/discussion/172458).\n\nAlso, the notebook for EfficientDet from @shonenkov is available [here](https://www.kaggle.com/shonenkov/training-efficientdet).\n\nSo I prepared [the training notebook](https://www.kaggle.com/osamurai/training-efficientdet-d5). \nThis is elementary notebook(version 0), so from now on I will implement some methods(e.g. other augmentation technique, Pseudo-Labeling and other if any)\n\nYou cannot use the shonenkov's notebook for this competion straightforward.\nSome modification are needed( e.g. dependencies such as albumentation and efficientdet libraries for me.)\n\nI am also preparing inference notebook in this weekend.\nI will release it soon.\n\n\nP.S.\nI made a mistake..\nI created the Notebook with pytorch.\nSo I will create tensorflow version.", "votes": null }, { "id": "1597301", "postDate": "11/27/2021 12:10:12", "content": "

Great Notebook, @osamurai, I was in a dilemma that do we have to use only Tensorflow. I thought using pytorch is also ok. Although I saw this in the Rule tab, Use TensorFlow 2.X on GPU

", "rawMarkdown": "Great Notebook, @osamurai, I was in a dilemma that do we have to use only Tensorflow. I thought using pytorch is also ok. Although I saw this in the `Rule` tab, `Use TensorFlow 2.X on GPU`", "votes": null }, { "id": "1597308", "postDate": "11/27/2021 12:13:06", "content": "

I think that's only if you want to be eligible for the TensorFlow Performance Prizes

", "rawMarkdown": "I think that's only if you want to be eligible for the TensorFlow Performance Prizes", "votes": null }, { "id": "1597328", "postDate": "11/27/2021 12:26:24", "content": "

Oh, I see thank you so much😇 for clearing the doubt. That's great, using only TensorFlow could be trouble.

", "rawMarkdown": "Oh, I see thank you so much😇 for clearing the doubt. That's great, using only TensorFlow could be trouble.", "votes": null }, { "id": "1597824", "postDate": "11/28/2021 00:55:08", "content": "
\n

I think that's only if you want to be eligible for the TensorFlow Performance Prizes

\n
\n

Thank you very much.

", "rawMarkdown": "> I think that's only if you want to be eligible for the TensorFlow Performance Prizes\n\nThank you very much.", "votes": null } ], "comments": [ { "id": 1597301, "author_name": "soumya9977", "author_url": "", "post_date": "11/27/2021 12:10:12", "content": "

Great Notebook, @osamurai, I was in a dilemma that do we have to use only Tensorflow. I thought using pytorch is also ok. Although I saw this in the Rule tab, Use TensorFlow 2.X on GPU

", "votes": null, "replies": [ { "id": 1597308, "author_name": "anjum48", "author_url": "", "post_date": "11/27/2021 12:13:06", "content": "

I think that's only if you want to be eligible for the TensorFlow Performance Prizes

", "votes": null, "replies": [ { "id": 1597824, "author_name": "osamurai", "author_url": "", "post_date": "11/28/2021 00:55:08", "content": "
\n

I think that's only if you want to be eligible for the TensorFlow Performance Prizes

\n
\n

Thank you very much.

", "votes": null, "replies": [] } ] }, { "id": 1597328, "author_name": "soumya9977", "author_url": "", "post_date": "11/27/2021 12:26:24", "content": "

Oh, I see thank you so much😇 for clearing the doubt. That's great, using only TensorFlow could be trouble.

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1597064": "As always, I am not familier with the object detection model, so I surveyed briefly.\nThen I found [**EficientDet**](https://arxiv.org/abs/1911.09070).\n\nThis model seems good because in the last competition called \"[Global Wheat Detection](https://www.kaggle.com/c/global-wheat-detection)\", grand master Mr @tereka won 5th place with this model.\n\nHis discussion is available [here](https://www.kaggle.com/c/global-wheat-detection/discussion/172458).\n\nAlso, the notebook for EfficientDet from @shonenkov is available [here](https://www.kaggle.com/shonenkov/training-efficientdet).\n\nSo I prepared [the training notebook](https://www.kaggle.com/osamurai/training-efficientdet-d5). \nThis is elementary notebook(version 0), so from now on I will implement some methods(e.g. other augmentation technique, Pseudo-Labeling and other if any)\n\nYou cannot use the shonenkov's notebook for this competion straightforward.\nSome modification are needed( e.g. dependencies such as albumentation and efficientdet libraries for me.)\n\nI am also preparing inference notebook in this weekend.\nI will release it soon.\n\n\nP.S.\nI made a mistake..\nI created the Notebook with pytorch.\nSo I will create tensorflow version.", "1597301": "Great Notebook, @osamurai, I was in a dilemma that do we have to use only Tensorflow. I thought using pytorch is also ok. Although I saw this in the `Rule` tab, `Use TensorFlow 2.X on GPU`", "1597308": "I think that's only if you want to be eligible for the TensorFlow Performance Prizes", "1597328": "Oh, I see thank you so much😇 for clearing the doubt. That's great, using only TensorFlow could be trouble.", "1597824": "> I think that's only if you want to be eligible for the TensorFlow Performance Prizes\n\nThank you very much." }, "source": "meta" }