Dear Organizer,
I have noticed that some sequences are missing JSON files (e.g., train/Seq_094_0037674 has only 21 JSON files). I might be missing some information, but is this an intentional design, or is there an issue with the dataset?
Created by Shunsuke Kikuchi ne_gi_chi__ That makes sense, thanks for the clarification! Hi ne\_gi\_chi__,
This is intentional design. Images that have neither visible folds or tools do not have a JSON file:
- You will realise that Seq094\_0037674 has 21 JSON files, but still has 40 segmentation maps (/segm\_maps)
- The image/mask indices that do not have a JSON file correspond to segmentation masks that are completely black (they are all background). If you look at the corresponding RGB images, you'll realise that this corresponds to occluded frames within the clip where the image appears all in an orange colour. The endoscope is very close to the colon wall and you can't see anything else. This is a "negative sample" where your predicted segmentation should classify the entire image as background.
I hope this is clear?
best
Francisco