I've trained models to predict coordinates of all levels of the spine needed for this competition but I haven't figured out a way to use them to improve score. I've tried adding masks around the predicted x, y coords as channels (6 channel image of the form [original_image, mask1, mask2, mask3, mask4, mask5]) and also cropping out each level and passing in 5 channel image with each level [level1_crop, level2_crop, …, level5_crop] but no improvement in CV. Maybe the model has already learned these features without needing coordinates. Has anyone used the coordinate data to improve their score?
", "messages": [ { "id": "2857320", "postDate": "06/05/2024 19:58:26", "content": "I've trained models to predict coordinates of all levels of the spine needed for this competition but I haven't figured out a way to use them to improve score. I've tried adding masks around the predicted x, y coords as channels (6 channel image of the form [original_image, mask1, mask2, mask3, mask4, mask5]) and also cropping out each level and passing in 5 channel image with each level [level1_crop, level2_crop, …, level5_crop] but no improvement in CV. Maybe the model has already learned these features without needing coordinates. Has anyone used the coordinate data to improve their score?
", "rawMarkdown": "I've trained models to predict coordinates of all levels of the spine needed for this competition but I haven't figured out a way to use them to improve score. I've tried adding masks around the predicted x, y coords as channels (6 channel image of the form [original_image, mask1, mask2, mask3, mask4, mask5]) and also cropping out each level and passing in 5 channel image with each level [level1_crop, level2_crop, ..., level5_crop] but no improvement in CV. Maybe the model has already learned these features without needing coordinates. Has anyone used the coordinate data to improve their score?", "votes": null }, { "id": "2858872", "postDate": "06/06/2024 17:31:58", "content": "I will try this and tell you. Are you training multiple models for each condition?
", "rawMarkdown": "I will try this and tell you. Are you training multiple models for each condition?", "votes": null }, { "id": "2871583", "postDate": "06/14/2024 09:23:42", "content": "In the test set there are no instance position where the stenosis can be found unlike in the train label coordinate csv file, how I go about this.
", "rawMarkdown": "In the test set there are no instance position where the stenosis can be found unlike in the train label coordinate csv file, how I go about this.", "votes": null }, { "id": "2871639", "postDate": "06/14/2024 10:03:44", "content": "train models to predict it or just discard it
", "rawMarkdown": "train models to predict it or just discard it", "votes": null }, { "id": "2871690", "postDate": "06/14/2024 10:44:39", "content": "What kinds of masks do you use? Have you ever thought of the shift invariance in CNN model?
", "rawMarkdown": "What kinds of masks do you use? Have you ever thought of the shift invariance in CNN model?", "votes": null }, { "id": "2872436", "postDate": "06/14/2024 19:07:56", "content": "please cold you walk me throught the process of makign predictions on the test images dataset
", "rawMarkdown": "please cold you walk me throught the process of makign predictions on the test images dataset", "votes": null }, { "id": "2877248", "postDate": "06/18/2024 09:03:20", "content": "You can either train a segmentation model by creating masks using x and y or, you can train an object detection model by creating bounding boxes using x and y
", "rawMarkdown": "You can either train a segmentation model by creating masks using x and y or, you can train an object detection model by creating bounding boxes using x and y", "votes": null }, { "id": "2877332", "postDate": "06/18/2024 09:52:16", "content": "the question is that the masks we created may not be accurate enough. any idea?
", "rawMarkdown": "the question is that the masks we created may not be accurate enough. any idea?", "votes": null }, { "id": "2878513", "postDate": "06/19/2024 02:47:36", "content": "That's the challenge. Am experimenting with different ranges for the radius, here is how am trying to create rectangular or circular masks; (x-radius, y-radius), (x+radius, y+radius)
", "rawMarkdown": "That's the challenge. Am experimenting with different ranges for the radius, here is how am trying to create rectangular or circular masks; (x-radius, y-radius), (x+radius, y+radius)", "votes": null }, { "id": "2878979", "postDate": "06/19/2024 09:48:01", "content": "sounds good. maybe we can also try different shapes of masks
", "rawMarkdown": "sounds good. maybe we can also try different shapes of masks", "votes": null }, { "id": "2883456", "postDate": "06/22/2024 02:13:54", "content": "I tried to use the coordinate to make a frame as ROI. but i later realized that coordinates are not provided in test set.
", "rawMarkdown": "I tried to use the coordinate to make a frame as ROI. but i later realized that coordinates are not provided in test set.", "votes": null } ], "comments": [ { "id": 2858872, "author_name": "manaidu", "author_url": "", "post_date": "06/06/2024 17:31:58", "content": "I will try this and tell you. Are you training multiple models for each condition?
", "votes": null, "replies": [] }, { "id": 2871583, "author_name": "theexaltedone", "author_url": "", "post_date": "06/14/2024 09:23:42", "content": "In the test set there are no instance position where the stenosis can be found unlike in the train label coordinate csv file, how I go about this.
", "votes": null, "replies": [ { "id": 2871639, "author_name": "kidougen", "author_url": "", "post_date": "06/14/2024 10:03:44", "content": "train models to predict it or just discard it
", "votes": null, "replies": [ { "id": 2872436, "author_name": "theexaltedone", "author_url": "", "post_date": "06/14/2024 19:07:56", "content": "please cold you walk me throught the process of makign predictions on the test images dataset
", "votes": null, "replies": [ { "id": 2877248, "author_name": "samu2505", "author_url": "", "post_date": "06/18/2024 09:03:20", "content": "You can either train a segmentation model by creating masks using x and y or, you can train an object detection model by creating bounding boxes using x and y
", "votes": null, "replies": [ { "id": 2877332, "author_name": "kidougen", "author_url": "", "post_date": "06/18/2024 09:52:16", "content": "the question is that the masks we created may not be accurate enough. any idea?
", "votes": null, "replies": [ { "id": 2878513, "author_name": "samu2505", "author_url": "", "post_date": "06/19/2024 02:47:36", "content": "That's the challenge. Am experimenting with different ranges for the radius, here is how am trying to create rectangular or circular masks; (x-radius, y-radius), (x+radius, y+radius)
", "votes": null, "replies": [ { "id": 2878979, "author_name": "kidougen", "author_url": "", "post_date": "06/19/2024 09:48:01", "content": "sounds good. maybe we can also try different shapes of masks
", "votes": null, "replies": [] } ] } ] } ] } ] } ] } ] }, { "id": 2871690, "author_name": "shipuxu", "author_url": "", "post_date": "06/14/2024 10:44:39", "content": "What kinds of masks do you use? Have you ever thought of the shift invariance in CNN model?
", "votes": null, "replies": [] }, { "id": 2883456, "author_name": "jackysywk", "author_url": "", "post_date": "06/22/2024 02:13:54", "content": "I tried to use the coordinate to make a frame as ROI. but i later realized that coordinates are not provided in test set.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2857320": "I've trained models to predict coordinates of all levels of the spine needed for this competition but I haven't figured out a way to use them to improve score. I've tried adding masks around the predicted x, y coords as channels (6 channel image of the form [original_image, mask1, mask2, mask3, mask4, mask5]) and also cropping out each level and passing in 5 channel image with each level [level1_crop, level2_crop, ..., level5_crop] but no improvement in CV. Maybe the model has already learned these features without needing coordinates. Has anyone used the coordinate data to improve their score?", "2858872": "I will try this and tell you. Are you training multiple models for each condition?", "2871583": "In the test set there are no instance position where the stenosis can be found unlike in the train label coordinate csv file, how I go about this.", "2871639": "train models to predict it or just discard it", "2871690": "What kinds of masks do you use? Have you ever thought of the shift invariance in CNN model?", "2872436": "please cold you walk me throught the process of makign predictions on the test images dataset", "2877248": "You can either train a segmentation model by creating masks using x and y or, you can train an object detection model by creating bounding boxes using x and y", "2877332": "the question is that the masks we created may not be accurate enough. any idea?", "2878513": "That's the challenge. Am experimenting with different ranges for the radius, here is how am trying to create rectangular or circular masks; (x-radius, y-radius), (x+radius, y+radius)", "2878979": "sounds good. maybe we can also try different shapes of masks", "2883456": "I tried to use the coordinate to make a frame as ROI. but i later realized that coordinates are not provided in test set." }, "source": "meta" }