{ "id": 246586, "title": "[simple baseline]efnb7 + yolov5 LB 0.595 ", "url": "/competitions/siim-covid19-detection/discussion/246586", "author_name": "Alien", "post_date": "2021-06-16T04:12:33.491000", "votes": 58, "comment_count": 10, "views": 0, "content": "

train_study: https://www.kaggle.com/h053473666/siim-covid19-efnb7-train-study
\ntrain_image: https://www.kaggle.com/h053473666/siim-cov19-yolov5-train
\ntrain_2class: https://www.kaggle.com/h053473666/siim-covid19-efnb7-train-fold0-5-2class
\ninfer_study(0.438): https://www.kaggle.com/h053473666/siim-cov19-efnb7-infer-study
\ninfer_all(0.595): https://www.kaggle.com/h053473666/siim-cov19-efnb7-yolov5-infer

\n

The three main predictions of this competition: study level classification prediction, image level object detection and image level two classification prediction.

", "messages": [ { "id": 1351073, "postDate": "2021-06-16T04:12:33.493Z", "content": "

train_study: https://www.kaggle.com/h053473666/siim-covid19-efnb7-train-study
\ntrain_image: https://www.kaggle.com/h053473666/siim-cov19-yolov5-train
\ntrain_2class: https://www.kaggle.com/h053473666/siim-covid19-efnb7-train-fold0-5-2class
\ninfer_study(0.438): https://www.kaggle.com/h053473666/siim-cov19-efnb7-infer-study
\ninfer_all(0.595): https://www.kaggle.com/h053473666/siim-cov19-efnb7-yolov5-infer

\n

The three main predictions of this competition: study level classification prediction, image level object detection and image level two classification prediction.

", "rawMarkdown": "train_study: https://www.kaggle.com/h053473666/siim-covid19-efnb7-train-study\ntrain_image: https://www.kaggle.com/h053473666/siim-cov19-yolov5-train\ntrain_2class: https://www.kaggle.com/h053473666/siim-covid19-efnb7-train-fold0-5-2class\ninfer_study(0.438): https://www.kaggle.com/h053473666/siim-cov19-efnb7-infer-study\ninfer_all(0.595): https://www.kaggle.com/h053473666/siim-cov19-efnb7-yolov5-infer\n\nThe three main predictions of this competition: study level classification prediction, image level object detection and image level two classification prediction.", "votes": 58 }, { "id": 1351696, "postDate": "2021-06-16T14:28:05.593Z", "content": "

What is the difference between \"train_2class\" and \"train_study\"?

", "rawMarkdown": "What is the difference between \"train_2class\" and \"train_study\"?", "votes": 3, "replies": [ { "id": 1351765, "postDate": "2021-06-16T15:26:33.197Z", "content": "

train_study:f'negative {negative} 0 0 1 1 typical {typical} 0 0 1 1 indeterminate {indeterminate} 0 0 1 1 atypical {atypical} 0 0 1 1'

\n

train_2class:f'none {none} 0 0 1 1'

\n

One is to classify Negative Typical Indeterminate Atypical. The other is to classify none. Because none and Negative are not exactly the same. So I add a model to predict it.

", "rawMarkdown": "train_study:`f'negative {negative} 0 0 1 1 typical {typical} 0 0 1 1 indeterminate {indeterminate} 0 0 1 1 atypical {atypical} 0 0 1 1'`\n\ntrain_2class:`f'none {none} 0 0 1 1'`\n\nOne is to classify Negative Typical Indeterminate Atypical. The other is to classify none. Because none and Negative are not exactly the same. So I add a model to predict it.\n", "votes": 7 }, { "id": 1365460, "postDate": "2021-06-25T18:38:36.230Z", "content": "

So you only do the bbox inference on the images that don't have the 'none' label?

", "rawMarkdown": "So you only do the bbox inference on the images that don't have the 'none' label?" } ] }, { "id": 1351180, "postDate": "2021-06-16T05:51:52.077Z", "content": "

There will be a new starting point for this competation. Fortunately, we still have enough time to try more methods on this baseline.

", "rawMarkdown": "There will be a new starting point for this competation. Fortunately, we still have enough time to try more methods on this baseline.", "votes": 1, "replies": [ { "id": 1351221, "postDate": "2021-06-16T06:47:27.437Z", "content": "

my estimation at the end of the competition:

\n

for public LB
\ngold cut off : 0.70~0.75
\n top1 : 0.80~0.85

\n

public LB is a different story

", "rawMarkdown": "my estimation at the end of the competition:\n\nfor public LB\ngold cut off : 0.70~0.75\n top1 : 0.80~0.85\n\npublic LB is a different story", "votes": 3 }, { "id": 1351227, "postDate": "2021-06-16T06:52:57.323Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 1351325, "postDate": "2021-06-16T08:09:09.537Z", "content": "

Hello. Thanks for your work - I've learned a lot. But I still have a question - what is the 2nd level of classification for?

", "rawMarkdown": "Hello. Thanks for your work - I've learned a lot. But I still have a question - what is the 2nd level of classification for?", "votes": 2, "replies": [ { "id": 1351356, "postDate": "2021-06-16T08:34:51.060Z", "content": "

opacity uses the object detection model to get the confidence value and bbox. none uses the classification model to get the confidence value.

", "rawMarkdown": "`opacity` uses the object detection model to get the confidence value and bbox. `none` uses the classification model to get the confidence value.", "votes": 3 }, { "id": 1351367, "postDate": "2021-06-16T08:51:56.847Z", "content": "

What if the values ​​at the exploration level and at the image level do not match? If we say negative at the research level and boxes at the image level? Isn't it better to weed out the negative and find boxes in the remaining images?

", "rawMarkdown": "What if the values ​​at the exploration level and at the image level do not match? If we say negative at the research level and boxes at the image level? Isn't it better to weed out the negative and find boxes in the remaining images?", "votes": 1 }, { "id": 1351372, "postDate": "2021-06-16T09:00:20.427Z", "content": "

Because opacity does not necessarily have pneumonia. Without opacity, there is not necessarily no pneumonia. For me, negative and (opacity, none) are different data.

", "rawMarkdown": "Because opacity does not necessarily have pneumonia. Without opacity, there is not necessarily no pneumonia. For me, negative and (opacity, none) are different data.", "votes": 1 } ] } ], "comments": [ { "id": 1351696, "author_name": "erkut", "author_url": "", "post_date": "2021-06-16T14:28:05.593000", "content": "

What is the difference between \"train_2class\" and \"train_study\"?

", "votes": 3, "replies": [ { "id": 1351765, "author_name": "Alien", "author_url": "", "post_date": "2021-06-16T15:26:33.197000", "content": "

train_study:f'negative {negative} 0 0 1 1 typical {typical} 0 0 1 1 indeterminate {indeterminate} 0 0 1 1 atypical {atypical} 0 0 1 1'

\n

train_2class:f'none {none} 0 0 1 1'

\n

One is to classify Negative Typical Indeterminate Atypical. The other is to classify none. Because none and Negative are not exactly the same. So I add a model to predict it.

", "votes": 7, "replies": [] }, { "id": 1365460, "author_name": "Adriano Passos", "author_url": "", "post_date": "2021-06-25T18:38:36.230000", "content": "

So you only do the bbox inference on the images that don't have the 'none' label?

", "votes": 0, "replies": [] } ] }, { "id": 1351180, "author_name": "Matthew Wu", "author_url": "", "post_date": "2021-06-16T05:51:52.077000", "content": "

There will be a new starting point for this competation. Fortunately, we still have enough time to try more methods on this baseline.

", "votes": 1, "replies": [ { "id": 1351221, "author_name": "hengck23", "author_url": "", "post_date": "2021-06-16T06:47:27.437000", "content": "

my estimation at the end of the competition:

\n

for public LB
\ngold cut off : 0.70~0.75
\n top1 : 0.80~0.85

\n

public LB is a different story

", "votes": 3, "replies": [] }, { "id": 1351227, "author_name": "", "author_url": "", "post_date": "2021-06-16T06:52:57.323000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 1351325, "author_name": "Zaakcii Ru", "author_url": "", "post_date": "2021-06-16T08:09:09.537000", "content": "

Hello. Thanks for your work - I've learned a lot. But I still have a question - what is the 2nd level of classification for?

", "votes": 2, "replies": [ { "id": 1351356, "author_name": "Alien", "author_url": "", "post_date": "2021-06-16T08:34:51.060000", "content": "

opacity uses the object detection model to get the confidence value and bbox. none uses the classification model to get the confidence value.

", "votes": 3, "replies": [] }, { "id": 1351367, "author_name": "Zaakcii Ru", "author_url": "", "post_date": "2021-06-16T08:51:56.847000", "content": "

What if the values ​​at the exploration level and at the image level do not match? If we say negative at the research level and boxes at the image level? Isn't it better to weed out the negative and find boxes in the remaining images?

", "votes": 1, "replies": [] }, { "id": 1351372, "author_name": "Alien", "author_url": "", "post_date": "2021-06-16T09:00:20.427000", "content": "

Because opacity does not necessarily have pneumonia. Without opacity, there is not necessarily no pneumonia. For me, negative and (opacity, none) are different data.

", "votes": 1, "replies": [] } ] } ], "raw_markdown_by_id": { "1351073": "train_study: https://www.kaggle.com/h053473666/siim-covid19-efnb7-train-study\ntrain_image: https://www.kaggle.com/h053473666/siim-cov19-yolov5-train\ntrain_2class: https://www.kaggle.com/h053473666/siim-covid19-efnb7-train-fold0-5-2class\ninfer_study(0.438): https://www.kaggle.com/h053473666/siim-cov19-efnb7-infer-study\ninfer_all(0.595): https://www.kaggle.com/h053473666/siim-cov19-efnb7-yolov5-infer\n\nThe three main predictions of this competition: study level classification prediction, image level object detection and image level two classification prediction.", "1351696": "What is the difference between \"train_2class\" and \"train_study\"?", "1351180": "There will be a new starting point for this competation. Fortunately, we still have enough time to try more methods on this baseline.", "1351325": "Hello. Thanks for your work - I've learned a lot. But I still have a question - what is the 2nd level of classification for?" } }