Challenge Dataset¶

CholecT50¶

CholecT50 is the first and largest endoscopic video dataset annotated with surgical action triplet labels for triplet recognition and detection. Its annotations include labels for instruments, verb, targets, phases, triplets, and other micro labels. The CholecT50 dataset is the official dataset for CholecTriplet 2022 challenge.

Training set ¶

1. The training set for the challenge is CholecT45 dataset consisting of 45 videos of CholecT50.
2. Use the download request form to obtain the CholecT45 dataset.
3. Only binary presence labels are provided for the training data.
4. No bounding box is provided to encourage localization by weak supervision in this challenge.
5. Bounding boxes (with box-triplet pairing) for some sample images to give a cue about the region of localization is provided here. It also include a sample code to visualize the images and labels.

For more information about the dataset, visit our GitHub repo.

Validation set ¶

1. Short video clips (sliced videos, part of CholecT45) are provided for self-validation.
2. Annotations = binary presence labels + spatial bounding boxes
3. Bounding box labels are outsourced from m2cai16-tool-location dataset.
4. Box-triplet pairing labels are be provided for the validation data.
5. Download the validation data here (also provided within the self-validator). Please note, that publication on this validation set is not allowed until a joint publication on the challenge methods and results is released.

Testing set¶

1. The testing set for the challenge is the remaining 5 videos of CholecT50 which are private.
2. Annotations = binary presence labels + spatial bounding boxes
3. The video images are fully annotated with bounding boxes and box-triplet pairing information.
4. The testing set will remain private throughout the duration of the challenge.

The data annotation protocol is explained here.

Happy challenge!