We have used the Cholec80 dataset which contains 80 videos of cholecystectomy procedures (surgical removal of the gallbladder) from the University Hospital of Strasbourg/IRCAD, France. The first forty videos are the fine-tuning subset from [1] and the remaining forms the evaluation subset from [1]. The videos have been recorded at 25 fps. The two types of annotations provided in the dataset are explained below:

Tool Annotations

For all 80 videos, the dataset contains tool annotation file for each video, which contains 8 columns. The first row of each file is the header. The first column contains the frame indexes (0 indexed). The other columns contain labels where 0 indicates absence of a tool and 1 indicates presence of a tool.

The dataset contains tool annotations which are generated at 1annotation per 25 frames. They indicate presence of seven tools namely:

1. Grasper

2. Bipolar

3. Hook

4. Scissors

5. Clipper

6. Irrigator

7. Specimen bag

Phase Annotations

For all 80 videos, the dataset contains phase annotation file for each video, which contains 2 columns. The first column corresponds to frame indexes (0 indexed) and the second column corresponds to phase label. The first row in the file contains header. These annotations help in visualizing phases.

All the frames are annotated with 8 defined phases namely,

1. Preparation

2. Calot triangle dissection

3. Clipping and cutting

4. Gallbladder dissection

5. Gallbladder packaging

6. Cleaning and coagulation

7. Gallbladder retraction