So this is my first ever attempt at a \"featured competition\", I have joined this competition with the goal of learn the most, but as of now I just have learned a bit about the data and object detection, I have not previously worked with CNN but I have a good understanding of how they work. I was wondering what are the approaches you guys take for these competition in general, I’m trying to figure out the best approach for tackling competitions like this, as I have only particpated in tabular competitions, where I would:
\n 1. Understand the data,
\n 2. Perform feature engineering,
\n 3. Build a pipeline, and
\n 4. Fit a model.
My questions are :
\nWhat does the typical process look like for featured competitions (like those involving computer vision, object detection, etc.)?
\nHow do you prepare for competitions?
\nHow do you plan for a competition if you’re not already proficient in the required techniques?
\nThank you.
", "messages": [ { "id": "3062442", "postDate": "12/03/2024 15:32:59", "content": "So this is my first ever attempt at a \"featured competition\", I have joined this competition with the goal of learn the most, but as of now I just have learned a bit about the data and object detection, I have not previously worked with CNN but I have a good understanding of how they work. I was wondering what are the approaches you guys take for these competition in general, I’m trying to figure out the best approach for tackling competitions like this, as I have only particpated in tabular competitions, where I would:
\n 1. Understand the data,
\n 2. Perform feature engineering,
\n 3. Build a pipeline, and
\n 4. Fit a model.
My questions are :
\nWhat does the typical process look like for featured competitions (like those involving computer vision, object detection, etc.)?
\nHow do you prepare for competitions?
\nHow do you plan for a competition if you’re not already proficient in the required techniques?
\nThank you.
", "rawMarkdown": "So this is my first ever attempt at a \"featured competition\", I have joined this competition with the goal of learn the most, but as of now I just have learned a bit about the data and object detection, I have not previously worked with CNN but I have a good understanding of how they work. I was wondering what are the approaches you guys take for these competition in general, I’m trying to figure out the best approach for tackling competitions like this, as I have only particpated in tabular competitions, where I would:\n 1. Understand the data,\n 2. Perform feature engineering,\n 3. Build a pipeline, and\n 4. Fit a model.\n\nMy questions are :\n1. What does the typical process look like for featured competitions (like those involving computer vision, object detection, etc.)?\n - How do you guys approach the problem from start to finish?\n\n2. How do you prepare for competitions?\n - Do you read research papers to understand the domain or task?\n - Do you analyze and learn from other people’s notebooks or code?\n\n3. How do you plan for a competition if you’re not already proficient in the required techniques?\n - For example, if you were new to object detection, how would you go about learning it?\n - What steps would you take to become skilled enough to achieve a competitive position on the leaderboard?\n\nThank you.", "votes": null }, { "id": "3063016", "postDate": "12/04/2024 05:10:11", "content": "Start here to understand the data (I may be a little biased on this one):
\nhttps://www.kaggle.com/code/davidlist/experiment-ts-6-4-visualization
Go through this notebook line by line and understand what it does. Submit it, and then figure out ways to make it better:
\nhttps://www.kaggle.com/code/fnands/baseline-unet-train-submit
Visit this site:
\nhttps://cryoetdataportal.czscience.com/competition
\nAnd this:
\nhttps://github.com/czimaginginstitute/2024_czii_mlchallenge_notebooks
\nAnd this:
\nhttps://cryoetdataportal.czscience.com/datasets/10441
Read the discussion posts. Pay special attention to any posts by @hengck23. For instance:
\nhttps://www.kaggle.com/competitions/czii-cryo-et-object-identification/discussion/547350
Hope this helps!
", "rawMarkdown": "Start here to understand the data (I may be a little biased on this one):\n[https://www.kaggle.com/code/davidlist/experiment-ts-6-4-visualization](https://www.kaggle.com/code/davidlist/experiment-ts-6-4-visualization)\n\nGo through this notebook line by line and understand what it does. Submit it, and then figure out ways to make it better:\nhttps://www.kaggle.com/code/fnands/baseline-unet-train-submit\n\nVisit this site:\n[https://cryoetdataportal.czscience.com/competition](https://cryoetdataportal.czscience.com/competition)\nAnd this:\n[https://github.com/czimaginginstitute/2024_czii_mlchallenge_notebooks](https://github.com/czimaginginstitute/2024_czii_mlchallenge_notebooks)\nAnd this:\n[https://cryoetdataportal.czscience.com/datasets/10441](https://cryoetdataportal.czscience.com/datasets/10441)\n\nRead the discussion posts. Pay special attention to any posts by @hengck23. For instance:\n[https://www.kaggle.com/competitions/czii-cryo-et-object-identification/discussion/547350](https://www.kaggle.com/competitions/czii-cryo-et-object-identification/discussion/547350)\n\nHope this helps!", "votes": null }, { "id": "3063074", "postDate": "12/04/2024 06:13:53", "content": "Thank you so much @davidlist for this.
", "rawMarkdown": "Thank you so much @davidlist for this.", "votes": null }, { "id": "3063577", "postDate": "12/04/2024 16:16:06", "content": "Hi David, could you please share some knowledge on how to get data labels from the '10441' dataset
", "rawMarkdown": "Hi David, could you please share some knowledge on how to get data labels from the '10441' dataset", "votes": null }, { "id": "3063738", "postDate": "12/04/2024 20:06:45", "content": "I'm only just downloading it now for the first time. First impression is the directory structure is a bit different so understanding how to use it may be its own little project.
", "rawMarkdown": "I'm only just downloading it now for the first time. First impression is the directory structure is a bit different so understanding how to use it may be its own little project.", "votes": null }, { "id": "3064711", "postDate": "12/05/2024 23:28:55", "content": "I think this notebook goes a long way towards an answer:
\nhttps://www.kaggle.com/code/davidlist/ts-0-simulated-data-visualization/notebook
Or actually, just use this notebook. https://www.kaggle.com/code/davidlist/simulated-data-and-labels :)
", "rawMarkdown": "Or actually, just use this notebook. [https://www.kaggle.com/code/davidlist/simulated-data-and-labels](https://www.kaggle.com/code/davidlist/simulated-data-and-labels) :)", "votes": null }, { "id": "3067310", "postDate": "12/09/2024 06:30:06", "content": "thanks a lot.
", "rawMarkdown": "thanks a lot.", "votes": null }, { "id": "3087229", "postDate": "01/03/2025 09:00:58", "content": "Hi! Thank you for your advice. I've gone through the baseline U-Net model. Could you please share some suggestions or directions on how to further improve it?
", "rawMarkdown": "Hi! Thank you for your advice. I've gone through the baseline U-Net model. Could you please share some suggestions or directions on how to further improve it?", "votes": null }, { "id": "3087837", "postDate": "01/04/2025 00:06:10", "content": "There are quite a few bugs in that notebook that affect performance, but probably the most important exercise relates to understanding the concept behind a precision-recall curve and how to use that to optimize your f4 score. Doing that should get you well into the 0.500's.
", "rawMarkdown": "There are quite a few bugs in that notebook that affect performance, but probably the most important exercise relates to understanding the concept behind a precision-recall curve and how to use that to optimize your f4 score. Doing that should get you well into the 0.500's.", "votes": null } ], "comments": [ { "id": 3063016, "author_name": "davidlist", "author_url": "", "post_date": "12/04/2024 05:10:11", "content": "Start here to understand the data (I may be a little biased on this one):
\nhttps://www.kaggle.com/code/davidlist/experiment-ts-6-4-visualization
Go through this notebook line by line and understand what it does. Submit it, and then figure out ways to make it better:
\nhttps://www.kaggle.com/code/fnands/baseline-unet-train-submit
Visit this site:
\nhttps://cryoetdataportal.czscience.com/competition
\nAnd this:
\nhttps://github.com/czimaginginstitute/2024_czii_mlchallenge_notebooks
\nAnd this:
\nhttps://cryoetdataportal.czscience.com/datasets/10441
Read the discussion posts. Pay special attention to any posts by @hengck23. For instance:
\nhttps://www.kaggle.com/competitions/czii-cryo-et-object-identification/discussion/547350
Hope this helps!
", "votes": null, "replies": [ { "id": 3063074, "author_name": "aquibahmad7", "author_url": "", "post_date": "12/04/2024 06:13:53", "content": "Thank you so much @davidlist for this.
", "votes": null, "replies": [] }, { "id": 3063577, "author_name": "stronggaylocksman", "author_url": "", "post_date": "12/04/2024 16:16:06", "content": "Hi David, could you please share some knowledge on how to get data labels from the '10441' dataset
", "votes": null, "replies": [ { "id": 3063738, "author_name": "davidlist", "author_url": "", "post_date": "12/04/2024 20:06:45", "content": "I'm only just downloading it now for the first time. First impression is the directory structure is a bit different so understanding how to use it may be its own little project.
", "votes": null, "replies": [] }, { "id": 3064711, "author_name": "davidlist", "author_url": "", "post_date": "12/05/2024 23:28:55", "content": "I think this notebook goes a long way towards an answer:
\nhttps://www.kaggle.com/code/davidlist/ts-0-simulated-data-visualization/notebook
Or actually, just use this notebook. https://www.kaggle.com/code/davidlist/simulated-data-and-labels :)
", "votes": null, "replies": [ { "id": 3067310, "author_name": "stronggaylocksman", "author_url": "", "post_date": "12/09/2024 06:30:06", "content": "thanks a lot.
", "votes": null, "replies": [] } ] } ] }, { "id": 3087229, "author_name": "biancochiu", "author_url": "", "post_date": "01/03/2025 09:00:58", "content": "Hi! Thank you for your advice. I've gone through the baseline U-Net model. Could you please share some suggestions or directions on how to further improve it?
", "votes": null, "replies": [ { "id": 3087837, "author_name": "davidlist", "author_url": "", "post_date": "01/04/2025 00:06:10", "content": "There are quite a few bugs in that notebook that affect performance, but probably the most important exercise relates to understanding the concept behind a precision-recall curve and how to use that to optimize your f4 score. Doing that should get you well into the 0.500's.
", "votes": null, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "3062442": "So this is my first ever attempt at a \"featured competition\", I have joined this competition with the goal of learn the most, but as of now I just have learned a bit about the data and object detection, I have not previously worked with CNN but I have a good understanding of how they work. I was wondering what are the approaches you guys take for these competition in general, I’m trying to figure out the best approach for tackling competitions like this, as I have only particpated in tabular competitions, where I would:\n 1. Understand the data,\n 2. Perform feature engineering,\n 3. Build a pipeline, and\n 4. Fit a model.\n\nMy questions are :\n1. What does the typical process look like for featured competitions (like those involving computer vision, object detection, etc.)?\n - How do you guys approach the problem from start to finish?\n\n2. How do you prepare for competitions?\n - Do you read research papers to understand the domain or task?\n - Do you analyze and learn from other people’s notebooks or code?\n\n3. How do you plan for a competition if you’re not already proficient in the required techniques?\n - For example, if you were new to object detection, how would you go about learning it?\n - What steps would you take to become skilled enough to achieve a competitive position on the leaderboard?\n\nThank you.", "3063016": "Start here to understand the data (I may be a little biased on this one):\n[https://www.kaggle.com/code/davidlist/experiment-ts-6-4-visualization](https://www.kaggle.com/code/davidlist/experiment-ts-6-4-visualization)\n\nGo through this notebook line by line and understand what it does. Submit it, and then figure out ways to make it better:\nhttps://www.kaggle.com/code/fnands/baseline-unet-train-submit\n\nVisit this site:\n[https://cryoetdataportal.czscience.com/competition](https://cryoetdataportal.czscience.com/competition)\nAnd this:\n[https://github.com/czimaginginstitute/2024_czii_mlchallenge_notebooks](https://github.com/czimaginginstitute/2024_czii_mlchallenge_notebooks)\nAnd this:\n[https://cryoetdataportal.czscience.com/datasets/10441](https://cryoetdataportal.czscience.com/datasets/10441)\n\nRead the discussion posts. Pay special attention to any posts by @hengck23. For instance:\n[https://www.kaggle.com/competitions/czii-cryo-et-object-identification/discussion/547350](https://www.kaggle.com/competitions/czii-cryo-et-object-identification/discussion/547350)\n\nHope this helps!", "3063074": "Thank you so much @davidlist for this.", "3063577": "Hi David, could you please share some knowledge on how to get data labels from the '10441' dataset", "3063738": "I'm only just downloading it now for the first time. First impression is the directory structure is a bit different so understanding how to use it may be its own little project.", "3064711": "I think this notebook goes a long way towards an answer:\n[https://www.kaggle.com/code/davidlist/ts-0-simulated-data-visualization/notebook](https://www.kaggle.com/code/davidlist/ts-0-simulated-data-visualization/notebook)", "3067164": "Or actually, just use this notebook. [https://www.kaggle.com/code/davidlist/simulated-data-and-labels](https://www.kaggle.com/code/davidlist/simulated-data-and-labels) :)", "3067310": "thanks a lot.", "3087229": "Hi! Thank you for your advice. I've gone through the baseline U-Net model. Could you please share some suggestions or directions on how to further improve it?", "3087837": "There are quite a few bugs in that notebook that affect performance, but probably the most important exercise relates to understanding the concept behind a precision-recall curve and how to use that to optimize your f4 score. Doing that should get you well into the 0.500's." }, "source": "meta" }