- Studies either contain a single image (the majority) or multiple images (up to 9)
- All of the images within the studies are also found in the image level dataset
- If a study contains multiple images, a maximum of one image and a minimum of zero images will have bounding boxes even though all of the images exist in the image level dataset.
- There are 2 typical scenarios that occur when multiple images are in a study:
- The images are nearly identical (relatively common - differences are blurriness/sharpness/lighting/rotation)(see the example(s) a & c below)
- The images are different from each other (relatively uncommon)(see the example b below)
- The images often have something wrong with them. This defect usually falls into one of the following categories
- poor quality, blurry, too-sharp, perspective transform, cropped poorly, exposure/brightness, skew, etc.
'머신러닝 > 캐글' 카테고리의 다른 글
finetuning시 big transfer and efficientnet v2 (0) | 2021.07.10 |
---|---|
캐글 전략 (0) | 2021.07.09 |
캐글 SIIM-FISABIO-RSNA COVID-19 Detection -심화 전략 (0) | 2021.07.02 |
캐글 SIIM-FISABIO-RSNA COVID-19 Detection - 베이스 전략 (0) | 2021.07.01 |