Would you rather like a pure torch notebook or detectron/mmdetection
\nImporting library like mm or detectron allows us to write less code, but it's so difficult to debug. Even if I fix bugs, I feel like I haven't learned anything because it totally depends on that library.
\nI wonder if it's more difficult to get good ranking in this kind of competion with pure pytorch(instance segmentation/object detection), since it looks like most of the player use library.
\nWhich one do you choose?
", "messages": [ { "id": "1574167", "postDate": "11/07/2021 09:44:55", "content": "Would you rather like a pure torch notebook or detectron/mmdetection
\nImporting library like mm or detectron allows us to write less code, but it's so difficult to debug. Even if I fix bugs, I feel like I haven't learned anything because it totally depends on that library.
\nI wonder if it's more difficult to get good ranking in this kind of competion with pure pytorch(instance segmentation/object detection), since it looks like most of the player use library.
\nWhich one do you choose?
", "rawMarkdown": "Would you rather like a [pure torch notebook](https://www.kaggle.com/rluethy/sartorius-torch-mask-r-cnn/notebook) or [detectron](https://www.kaggle.com/slawekbiel/positive-score-with-detectron-3-3-inference)/[mmdetection](https://www.kaggle.com/slawekbiel/positive-score-with-detectron-3-3-inference) \n\nImporting library like mm or detectron allows us to write less code, but it's so difficult to debug. Even if I fix bugs, I feel like I haven't learned anything because it totally depends on that library.\n\nI wonder if it's more difficult to get good ranking in this kind of competion with pure pytorch(instance segmentation/object detection), since it looks like most of the player use library.\n\nWhich one do you choose?", "votes": null }, { "id": "1574990", "postDate": "11/08/2021 04:39:13", "content": "I would start with Library then move to pure if needed.
\nTo get good ranking ,the tool is just part of the story.
The thing with libraries is that they get you off the ground much, much faster. I started with knowing nothing about segmentation and had a high scoring submission on day 5 of the competition just by following a tutorial and googling around. I would never have gotten there without having ready examples and code that does most of the things for me.
", "rawMarkdown": "The thing with libraries is that they get you off the ground much, much faster. I started with knowing nothing about segmentation and had a high scoring submission on day 5 of the competition just by following a tutorial and googling around. I would never have gotten there without having ready examples and code that does most of the things for me.", "votes": null }, { "id": "1576651", "postDate": "11/09/2021 11:17:11", "content": "I prefer to start with Library. 😄
", "rawMarkdown": "I prefer to start with Library. 😄", "votes": null }, { "id": "1576790", "postDate": "11/09/2021 13:15:14", "content": "I prefer library mmdetection, offers custom config file. Easy to add different augmentation like mosaic and mixup, mixed precision and other different methods. Will switch to pure torch when required.
", "rawMarkdown": "I prefer library mmdetection, offers custom config file. Easy to add different augmentation like mosaic and mixup, mixed precision and other different methods. Will switch to pure torch when required.", "votes": null } ], "comments": [ { "id": 1574990, "author_name": "faisalalsrheed", "author_url": "", "post_date": "11/08/2021 04:39:13", "content": "I would start with Library then move to pure if needed.
\nTo get good ranking ,the tool is just part of the story.
The thing with libraries is that they get you off the ground much, much faster. I started with knowing nothing about segmentation and had a high scoring submission on day 5 of the competition just by following a tutorial and googling around. I would never have gotten there without having ready examples and code that does most of the things for me.
", "votes": null, "replies": [] }, { "id": 1576651, "author_name": "kooaslansefat", "author_url": "", "post_date": "11/09/2021 11:17:11", "content": "I prefer to start with Library. 😄
", "votes": null, "replies": [] }, { "id": 1576790, "author_name": "mlneo07", "author_url": "", "post_date": "11/09/2021 13:15:14", "content": "I prefer library mmdetection, offers custom config file. Easy to add different augmentation like mosaic and mixup, mixed precision and other different methods. Will switch to pure torch when required.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1574167": "Would you rather like a [pure torch notebook](https://www.kaggle.com/rluethy/sartorius-torch-mask-r-cnn/notebook) or [detectron](https://www.kaggle.com/slawekbiel/positive-score-with-detectron-3-3-inference)/[mmdetection](https://www.kaggle.com/slawekbiel/positive-score-with-detectron-3-3-inference) \n\nImporting library like mm or detectron allows us to write less code, but it's so difficult to debug. Even if I fix bugs, I feel like I haven't learned anything because it totally depends on that library.\n\nI wonder if it's more difficult to get good ranking in this kind of competion with pure pytorch(instance segmentation/object detection), since it looks like most of the player use library.\n\nWhich one do you choose?", "1574990": "I would start with Library then move to pure if needed.\nTo get good ranking ,the tool is just part of the story.", "1575211": "The thing with libraries is that they get you off the ground much, much faster. I started with knowing nothing about segmentation and had a high scoring submission on day 5 of the competition just by following a tutorial and googling around. I would never have gotten there without having ready examples and code that does most of the things for me.", "1576651": "I prefer to start with Library. 😄", "1576790": "I prefer library mmdetection, offers custom config file. Easy to add different augmentation like mosaic and mixup, mixed precision and other different methods. Will switch to pure torch when required." }, "source": "meta" }