This competition seems to be very interesting to me. I wanted to ask competitors that is this compute-intensive competition (as Bristol Myers Squibb - Molecular Translation was in the past). Or it can be managed with limited computing power. I wanted to enter this comp. But others are going on as well, which require some compute power, so I cannot dedicate all of the compute to this comp.
", "messages": [ { "id": "1380529", "postDate": "07/08/2021 06:45:40", "content": "This competition seems to be very interesting to me. I wanted to ask competitors that is this compute-intensive competition (as Bristol Myers Squibb - Molecular Translation was in the past). Or it can be managed with limited computing power. I wanted to enter this comp. But others are going on as well, which require some compute power, so I cannot dedicate all of the compute to this comp.
", "rawMarkdown": "This competition seems to be very interesting to me. I wanted to ask competitors that is this compute-intensive competition (as **Bristol Myers Squibb - Molecular Translation** was in the past). Or it can be managed with limited computing power. I wanted to enter this comp. But others are going on as well, which require some compute power, so I cannot dedicate all of the compute to this comp.", "votes": null }, { "id": "1380852", "postDate": "07/08/2021 11:22:08", "content": "This competition is compute-intensive. DICOM chest x-ray images are large files (~3000x3000 pixels) that are usually stored in 16 bits. Also, some of the images are JPG encoded which requires external libraries and more CPU/GPU to decode.
", "rawMarkdown": "This competition is compute-intensive. DICOM chest x-ray images are large files (~3000x3000 pixels) that are usually stored in 16 bits. Also, some of the images are JPG encoded which requires external libraries and more CPU/GPU to decode.", "votes": null }, { "id": "1380895", "postDate": "07/08/2021 12:18:15", "content": "No, this is not a compute-extensive competition. You can just download one of the preprocessed jpg datasets posted here on the forums. Data is very small, models fit very fast. This is no comparison to Molecular competition.
", "rawMarkdown": "No, this is not a compute-extensive competition. You can just download one of the preprocessed jpg datasets posted here on the forums. Data is very small, models fit very fast. This is no comparison to Molecular competition.", "votes": null }, { "id": "1380900", "postDate": "07/08/2021 12:22:22", "content": "Thanks for the clarification @philippsinger …
", "rawMarkdown": "Thanks for the clarification @philippsinger ...", "votes": null }, { "id": "1380944", "postDate": "07/08/2021 12:56:35", "content": "I didn't compete in the Molecular comp, so I can't compare the two. But an exported set of JPGs in this comp (at full size) is still ~20G .. My crappy version of Alien's great yolo/effnet infer notebook (https://www.kaggle.com/h053473666/siim-cov19-efnb7-yolov5-infer) takes a couple hours of Kaggle GPU. I must be doing something wrong.
\nHow do you use a resized dataset without scaled BB data @philippsinger ?
", "rawMarkdown": "I didn't compete in the Molecular comp, so I can't compare the two. But an exported set of JPGs in this comp (at full size) is still ~20G .. My crappy version of Alien's great yolo/effnet infer notebook (https://www.kaggle.com/h053473666/siim-cov19-efnb7-yolov5-infer) takes a couple hours of Kaggle GPU. I must be doing something wrong.\n\nHow do you use a resized dataset without scaled BB data @philippsinger ?", "votes": null } ], "comments": [ { "id": 1380852, "author_name": "davidbroberts", "author_url": "", "post_date": "07/08/2021 11:22:08", "content": "This competition is compute-intensive. DICOM chest x-ray images are large files (~3000x3000 pixels) that are usually stored in 16 bits. Also, some of the images are JPG encoded which requires external libraries and more CPU/GPU to decode.
", "votes": null, "replies": [ { "id": 1380895, "author_name": "philippsinger", "author_url": "", "post_date": "07/08/2021 12:18:15", "content": "No, this is not a compute-extensive competition. You can just download one of the preprocessed jpg datasets posted here on the forums. Data is very small, models fit very fast. This is no comparison to Molecular competition.
", "votes": null, "replies": [] }, { "id": 1380900, "author_name": "atharvaingle", "author_url": "", "post_date": "07/08/2021 12:22:22", "content": "Thanks for the clarification @philippsinger …
", "votes": null, "replies": [] }, { "id": 1380944, "author_name": "davidbroberts", "author_url": "", "post_date": "07/08/2021 12:56:35", "content": "I didn't compete in the Molecular comp, so I can't compare the two. But an exported set of JPGs in this comp (at full size) is still ~20G .. My crappy version of Alien's great yolo/effnet infer notebook (https://www.kaggle.com/h053473666/siim-cov19-efnb7-yolov5-infer) takes a couple hours of Kaggle GPU. I must be doing something wrong.
\nHow do you use a resized dataset without scaled BB data @philippsinger ?
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1380529": "This competition seems to be very interesting to me. I wanted to ask competitors that is this compute-intensive competition (as **Bristol Myers Squibb - Molecular Translation** was in the past). Or it can be managed with limited computing power. I wanted to enter this comp. But others are going on as well, which require some compute power, so I cannot dedicate all of the compute to this comp.", "1380852": "This competition is compute-intensive. DICOM chest x-ray images are large files (~3000x3000 pixels) that are usually stored in 16 bits. Also, some of the images are JPG encoded which requires external libraries and more CPU/GPU to decode.", "1380895": "No, this is not a compute-extensive competition. You can just download one of the preprocessed jpg datasets posted here on the forums. Data is very small, models fit very fast. This is no comparison to Molecular competition.", "1380900": "Thanks for the clarification @philippsinger ...", "1380944": "I didn't compete in the Molecular comp, so I can't compare the two. But an exported set of JPGs in this comp (at full size) is still ~20G .. My crappy version of Alien's great yolo/effnet infer notebook (https://www.kaggle.com/h053473666/siim-cov19-efnb7-yolov5-infer) takes a couple hours of Kaggle GPU. I must be doing something wrong.\n\nHow do you use a resized dataset without scaled BB data @philippsinger ?" }, "source": "meta" }