I have good experience in segmentation however this is my first competition 🙃
\nIt is also not entirely clear to me how we send the result, but there is a mask. In the forecast line we have some kind of line. I'm assuming it's some sort of coded information.
\nWe have to send our neural network prediction on the test dataset. However, in the test folder we have only one picture. Do we really have to send a prediction of only one value?
\nIn the requirements for the competition, the notebook must be disconnected from the Internet. Do I understand correctly that, in fact, the competition requires that we use only pre-installed libraries and not download third-party repositories?
\nI will be immensely grateful for your help 💚💚💚
", "messages": [ { "id": "2295860", "postDate": "06/11/2023 10:17:01", "content": "I have good experience in segmentation however this is my first competition 🙃
\nIt is also not entirely clear to me how we send the result, but there is a mask. In the forecast line we have some kind of line. I'm assuming it's some sort of coded information.
\nWe have to send our neural network prediction on the test dataset. However, in the test folder we have only one picture. Do we really have to send a prediction of only one value?
\nIn the requirements for the competition, the notebook must be disconnected from the Internet. Do I understand correctly that, in fact, the competition requires that we use only pre-installed libraries and not download third-party repositories?
\nI will be immensely grateful for your help 💚💚💚
", "rawMarkdown": "**I have good experience in segmentation however this is my first competition 🙃**\n\n# Test data\n**It is also not entirely clear to me how we send the result, but there is a mask. In the forecast line we have some kind of line. I'm assuming it's some sort of coded information.**\n\n# Submission\n**We have to send our neural network prediction on the test dataset. However, in the test folder we have only one picture. Do we really have to send a prediction of only one value?**\n\n# Internet off in notebook\n**In the requirements for the competition, the notebook must be disconnected from the Internet. Do I understand correctly that, in fact, the competition requires that we use only pre-installed libraries and not download third-party repositories?**\n\nI will be immensely grateful for your help 💚💚💚", "votes": null }, { "id": "2295876", "postDate": "06/11/2023 10:30:59", "content": "\n\nIt is also not entirely clear to me how we send the result, but there is a mask. In the forecast line we have some kind of line. I'm assuming it's some sort of coded information.
\n
You can use the function given here to encode a binary mask into a compressed string for submission.
\n\n\nWe have to send our neural network prediction on the test dataset. However, in the test folder we have only one picture. Do we really have to send a prediction of only one value?
\n
You only have one test value available publicly, however when you submit the notebook, more images will be placed in that folder. It's basically so no one can see the final data before the end of the competition.
\n\n\nDo I understand correctly that, in fact, the competition requires that we use only pre-installed libraries and not download third-party repositories?
\n
While the internet is off, you can still use 3rd party libraries, you just need to install them as datasets. As an example, I do it here when I install pycocotools.
\nBasically it's so you don't secretly send the test data somewhere when you do inference.
Thank you so much. It helped me a lot 🙃
", "rawMarkdown": "Thank you so much. It helped me a lot 🙃", "votes": null }, { "id": "2300653", "postDate": "06/13/2023 10:35:33", "content": "\n\nBasically it's so you don't secretly send the test data somewhere when you do inference.
\n
Lets imagine the submission pipeline composed of 2 notebooks: the first one for training (it saves fitted model as dataset for further use) and the second one for prediction.
\nSo, the training notebook can use internet connection to install libraries and downloading pretrained models in a familiar way (without uploading all as datasets), and this constraint is applied only to the submission notebook?
Wow, This is exactly what I wanted to know now, thank you very much 😊
", "rawMarkdown": "Wow, This is exactly what I wanted to know now, thank you very much 😊", "votes": null }, { "id": "2300769", "postDate": "06/13/2023 12:33:07", "content": "I don't now if it is true, looking for answer from @yashvrdnjain @katherinegustilo
", "rawMarkdown": "I don't now if it is true, looking for answer from @yashvrdnjain @katherinegustilo", "votes": null }, { "id": "2300800", "postDate": "06/13/2023 13:00:41", "content": "Exactly. Only the submission notebook cannot access the internet.
\nFor the training one you can do whatever.
As an example: my training notebook uses teh internet (i.e. installs some packages with pip), and also build python wheels from the downloaded packages, which are then used in the inference notebook
", "rawMarkdown": "Exactly. Only the submission notebook cannot access the internet. \nFor the training one you can do whatever. \n\nAs an example: my [training notebook](https://www.kaggle.com/code/fnands/a-quick-yolov7-baseline) uses teh internet (i.e. installs some packages with pip), and also build python wheels from the downloaded packages, which are then used in the [inference notebook](https://www.kaggle.com/code/fnands/a-quick-yolov7-baseline-inference)", "votes": null }, { "id": "2300957", "postDate": "06/13/2023 14:31:15", "content": "Thanks friends. You help me so much 🙃
", "rawMarkdown": "Thanks friends. You help me so much 🙃", "votes": null } ], "comments": [ { "id": 2295876, "author_name": "fnands", "author_url": "", "post_date": "06/11/2023 10:30:59", "content": "\n\nIt is also not entirely clear to me how we send the result, but there is a mask. In the forecast line we have some kind of line. I'm assuming it's some sort of coded information.
\n
You can use the function given here to encode a binary mask into a compressed string for submission.
\n\n\nWe have to send our neural network prediction on the test dataset. However, in the test folder we have only one picture. Do we really have to send a prediction of only one value?
\n
You only have one test value available publicly, however when you submit the notebook, more images will be placed in that folder. It's basically so no one can see the final data before the end of the competition.
\n\n\nDo I understand correctly that, in fact, the competition requires that we use only pre-installed libraries and not download third-party repositories?
\n
While the internet is off, you can still use 3rd party libraries, you just need to install them as datasets. As an example, I do it here when I install pycocotools.
\nBasically it's so you don't secretly send the test data somewhere when you do inference.
Thank you so much. It helped me a lot 🙃
", "votes": null, "replies": [] }, { "id": 2300653, "author_name": "ogurtsov", "author_url": "", "post_date": "06/13/2023 10:35:33", "content": "\n\nBasically it's so you don't secretly send the test data somewhere when you do inference.
\n
Lets imagine the submission pipeline composed of 2 notebooks: the first one for training (it saves fitted model as dataset for further use) and the second one for prediction.
\nSo, the training notebook can use internet connection to install libraries and downloading pretrained models in a familiar way (without uploading all as datasets), and this constraint is applied only to the submission notebook?
Wow, This is exactly what I wanted to know now, thank you very much 😊
", "votes": null, "replies": [ { "id": 2300769, "author_name": "ogurtsov", "author_url": "", "post_date": "06/13/2023 12:33:07", "content": "I don't now if it is true, looking for answer from @yashvrdnjain @katherinegustilo
", "votes": null, "replies": [] } ] }, { "id": 2300800, "author_name": "fnands", "author_url": "", "post_date": "06/13/2023 13:00:41", "content": "Exactly. Only the submission notebook cannot access the internet.
\nFor the training one you can do whatever.
As an example: my training notebook uses teh internet (i.e. installs some packages with pip), and also build python wheels from the downloaded packages, which are then used in the inference notebook
", "votes": null, "replies": [] } ] } ] }, { "id": 2300957, "author_name": "", "author_url": "", "post_date": "06/13/2023 14:31:15", "content": "Thanks friends. You help me so much 🙃
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2295860": "**I have good experience in segmentation however this is my first competition 🙃**\n\n# Test data\n**It is also not entirely clear to me how we send the result, but there is a mask. In the forecast line we have some kind of line. I'm assuming it's some sort of coded information.**\n\n# Submission\n**We have to send our neural network prediction on the test dataset. However, in the test folder we have only one picture. Do we really have to send a prediction of only one value?**\n\n# Internet off in notebook\n**In the requirements for the competition, the notebook must be disconnected from the Internet. Do I understand correctly that, in fact, the competition requires that we use only pre-installed libraries and not download third-party repositories?**\n\nI will be immensely grateful for your help 💚💚💚", "2295876": "> It is also not entirely clear to me how we send the result, but there is a mask. In the forecast line we have some kind of line. I'm assuming it's some sort of coded information.\n\nYou can use the function [given here](https://www.kaggle.com/competitions/hubmap-hacking-the-human-vasculature/overview/evaluation) to encode a binary mask into a compressed string for submission. \n\n> We have to send our neural network prediction on the test dataset. However, in the test folder we have only one picture. Do we really have to send a prediction of only one value?\n\nYou only have one test value available publicly, however when you submit the notebook, more images will be placed in that folder. It's basically so no one can see the final data before the end of the competition. \n\n> Do I understand correctly that, in fact, the competition requires that we use only pre-installed libraries and not download third-party repositories?\n\nWhile the internet is off, you can still use 3rd party libraries, you just need to install them as datasets. As an example,[ I do it here ](https://www.kaggle.com/code/fnands/a-quick-yolov7-baseline-inference) when I install pycocotools. \nBasically it's so you don't secretly send the test data somewhere when you do inference.", "2295891": "Thank you so much. It helped me a lot 🙃", "2300653": ">Basically it's so you don't secretly send the test data somewhere when you do inference. \n\nLets imagine the submission pipeline composed of 2 notebooks: the first one for training (it saves fitted model as dataset for further use) and the second one for prediction. \nSo, the training notebook can use internet connection to install libraries and downloading pretrained models in a familiar way (without uploading all as datasets), and this constraint is applied only to the submission notebook?", "2300666": "Wow, This is exactly what I wanted to know now, thank you very much 😊", "2300769": "I don't now if it is true, looking for answer from @yashvrdnjain @katherinegustilo", "2300800": "Exactly. Only the submission notebook cannot access the internet. \nFor the training one you can do whatever. \n\nAs an example: my [training notebook](https://www.kaggle.com/code/fnands/a-quick-yolov7-baseline) uses teh internet (i.e. installs some packages with pip), and also build python wheels from the downloaded packages, which are then used in the [inference notebook](https://www.kaggle.com/code/fnands/a-quick-yolov7-baseline-inference)", "2300957": "Thanks friends. You help me so much 🙃" }, "source": "meta" }