Help needed: Regarding the paper appearing in the image, is the biomass in the area covered by the paper assumed to be zero by default?
", "messages": [ { "id": "3376872", "postDate": "12/15/2025 10:25:46", "content": "Help needed: Regarding the paper appearing in the image, is the biomass in the area covered by the paper assumed to be zero by default?
", "rawMarkdown": "Help needed: Regarding the paper appearing in the image, is the biomass in the area covered by the paper assumed to be zero by default?", "votes": null }, { "id": "3377037", "postDate": "12/15/2025 15:25:15", "content": "Pretty sure that is a bad assumption.
\nA better assumption would be that the images were never intended to be used for training a model. After all, vision models in 2017 could barely tell cats from dogs. The publication paper does make mention that photos were taken for quality control purposes, likewise the discussion about this competition suggests that 50% of the images they had needed to be discarded for various reasons. An image with a small amount of the id card would still be usable.
\nIn a discussion post someone shared an approach (clean_image) - there are a number of issues on the edge of many images to be handled. I am using a different approach but my view is that you do need to do something about the edges.
", "rawMarkdown": "Pretty sure that is a bad assumption. \n\nA better assumption would be that the images were never intended to be used for training a model. After all, vision models in 2017 could barely tell cats from dogs. The [publication paper](https://www.kaggle.com/competitions/csiro-biomass/discussion/613941) does make mention that photos were taken for quality control purposes, likewise the discussion about this competition suggests that 50% of the images they had needed to be discarded for various reasons. An image with a small amount of the id card would still be usable.\n\nIn a [discussion post](https://www.kaggle.com/competitions/csiro-biomass/discussion/651525#3376869) someone shared an approach (clean_image) - there are a number of issues on the edge of many images to be handled. I am using a different approach but my view is that you do need to do something about the edges.", "votes": null } ], "comments": [ { "id": 3377037, "author_name": "pcjimmmy", "author_url": "", "post_date": "12/15/2025 15:25:15", "content": "Pretty sure that is a bad assumption.
\nA better assumption would be that the images were never intended to be used for training a model. After all, vision models in 2017 could barely tell cats from dogs. The publication paper does make mention that photos were taken for quality control purposes, likewise the discussion about this competition suggests that 50% of the images they had needed to be discarded for various reasons. An image with a small amount of the id card would still be usable.
\nIn a discussion post someone shared an approach (clean_image) - there are a number of issues on the edge of many images to be handled. I am using a different approach but my view is that you do need to do something about the edges.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "3376872": "Help needed: Regarding the paper appearing in the image, is the biomass in the area covered by the paper assumed to be zero by default?", "3377037": "Pretty sure that is a bad assumption. \n\nA better assumption would be that the images were never intended to be used for training a model. After all, vision models in 2017 could barely tell cats from dogs. The [publication paper](https://www.kaggle.com/competitions/csiro-biomass/discussion/613941) does make mention that photos were taken for quality control purposes, likewise the discussion about this competition suggests that 50% of the images they had needed to be discarded for various reasons. An image with a small amount of the id card would still be usable.\n\nIn a [discussion post](https://www.kaggle.com/competitions/csiro-biomass/discussion/651525#3376869) someone shared an approach (clean_image) - there are a number of issues on the edge of many images to be handled. I am using a different approach but my view is that you do need to do something about the edges." }, "source": "meta" }