Clarification about Category 2 "other" label

Hi teams,

We wanted to clarify about the "other" class in category 2 labels. This class is essentially anything outside of the defined tasks that surgeons perform during training. There are some timeframes labelled as "other" in the csv labels file, however, anything unannotated within the videos are to be treated as "other" class (this will make "other" class one of the bigger ones in terms of data samples). The "other" class will be used in the final f1-evaluations as well.

Best, SurgVU Organizing Committee

Hi Aneeq,

Thanks for the clarification on the other class.

When I print the distinct task names from the CSV files in the dataset, I don’t see the other label being used for any of the case folders.

• Range of motion

• Rectal artery/vein

• Retraction and collision avoidance

• Skills application

• Suspensory ligaments

• Suturing

• Uterine horn

However, I observed that there are a few files with null values in the groundtruth_taskname column, as illustrated below (case_021->tasks.csv->linenr 10):

0,1,486.507643,1,718.647643,Suturing 1,1,5059.588643,1,5473.698643,Retraction and collision avoidance 2,1,6564.618643,1,7318.868643,Suturing 3,1,7700.368643,1,8927.638642999998,Uterine horn 4,1,9376.298643,1,10076.898643,Suturing 5,1,13113.198643,1,13476.498643,Range of motion 6,1,14239.498643,1,14860.598643,Suspensory ligaments 7,1,15462.698643,1,15774.398643,Suturing 8,1,16359.698643,1,17088.598643,Rectal artery/vein 9,1,17732.998643,1,17964.198643,Rectal artery/vein 10,2,2.3101129999995464,2,168.14064299999882,

Are you suggesting that rows in the tasks.csv file with no value in the groundtruth_taskname should be categorised as other? If this is not the correct interpretation, could you please provide a list of case folders where other is used as a label?

Thank you and Best regards, Bilal

Hi Bilal,

Yes, any timeframe not labelled is to be considered as "other" (along with any row that might have actually been labelled "other" or "null"). The labels provided have task labels per row and there are numerous timeframes in between tasks that are not indicated through a specific row or label in the csv files - all these timeframes are to be used as "other" class. Hope this clarifies it!

Best, SurgVU Org Team

Hi Aneeq,

Thank you for the clarification.

Just a quick follow-up: If a blank frame occurs between the start and end of a task, should we label it as “other” or should it carry the same label as the task segment?

Thanks again and best regards, Bilal

Hi Bilal,

Blank frames represent times when the endoscope might have gone out of body so it is up to the team on how to treat them (can be ignored as well to avoid training noise).

Best, SurgVU team

Thanks, Aneeq.

Will the test dataset used in the final phase contain any blank frames?

Hi Bilal,

Yes, the final test set may contain blank frames. However, keep in mind that blank frames are just a tiny fraction of the dataset, occurring only when the endoscope is out of the body.

Best, SurgVU Organizing Team