Frequently Asked Questions
--- Challenge ---
1. What is the Critical View of Safety (CVS)?
The Critical View of Safety is a maneuver performed during laparoscopic cholecystectomy that allows operators to safely proceed with the procedure.
2. Who can participate in the challenge?
Anyone can sign up to compete in the challenge (except members of the organizing labs), following the sign up instructions provided here.
3. Will the challenge submission still be open after the submission deadline?
No, only submissions made before the deadline will be considered eligible.
--- Registration ---
1. Why is my registration not approved yet?
For your registration to be approved, you must (1) click the join button in the top right corner on grandchallenge.org (2) Sign the participation agreement. Check the Instructions page for more details. If it has been more than 3 business days, please contact the organizers at info@cvschallenge.org.
2. Must every member of my team submit a signed participation agreement?
Yes, every team member must submit a signed participation agreement to receive access to the challenge dataset. You can find the agreement here. Upon submission, we will correlate the names that signed the agreement with those listed in the Challenge Team Form (make sure you are listed here!). Any submitting team member who has not submitted the signed agreement before the submission deadline may forfeit their eligibility to be considered as part of the submitting team and not be recognized for the submission.
--- CVS Challenge dataset ---
1. What validation split should I use?
Participants are free to subsample a portion of the provided training data to form an internal validation set for parameter tuning before submission.
2. Can I use private datasets to train my model?
No, the use of private datasets is strictly prohibited.
3. Can I use other public datasets to train my model?
Yes, participants are free to use any publicly available data to train and validate their submissions.
--- My Challenge Methods ---
1. Will the inference pipeline preserve the temporal information?
During testing, we will use an input setup that preserves the temporal frame order per video. Note that the frame rate during inference will be 1 fps, much lower than the frame rate of the original videos.
2. What is the frame rate for the test set?
1 frame per second.
3. Do I have to train my model using data sampled at 1 frame per second?
No, not at all. Participants are free to use any and all of the data provided within the scope of the challenge as well as any other public data.
4. Is the testing going to be an online prediction?
Yes, only use past and current frames to inform your predictions.
5. Can I tune my model at test time?
No, test-time tuning is not permitted.
6. The different challenge objectives seem to be conflicting, how do I prioritize what my method focuses on?
That's part of the challenge! Surgical Quality Assessments like the CVS assessment inherently present real-world difficulties like subjective interpretations, distribution shifts, etc... As we move toward real-world implementations of AI in surgery, we must balance different objectives such as being performant (subchallenge A), well-calibrated (subchallenge B), and robust (subchallenge C). You can choose to prioritize one or strike a balance between the different tasks.
7. Will the inference pipeline allow access to the metadata that was provided during training (e.g. use of ioc, icg, source location, etc.)?
No, the only inputs that will be available during test time will be image inputs extracted from the testing videos. While you can use these additional signals to train your model, your inference pipeline will need to essentially map a sequence of image inputs to a sequence of outputs and not rely on anything more.
--- Related Work ---
1. Are there public works on CVS assessment that you could point me to?
Of course, here are some published methods for CVS assessment [1], [2], [3], [4], [5].
2. Are there relevant datasets I should be looking at?
Here's one dataset for anatomy localization and CVS assessment as well as another public dataset for CVS assessment. Note, that while the target task may be the same (CVS assessment), the annotation protocols may vary and the use of these datasets needs to be carefully considered. Aside from this, several public benchmarks on various other related tasks for surgical video understanding do exist (here's a nice compilation) . We encourage participants to leverage these existing works into their submissions.
--- Submission ---
1. How do I submit my method?
Methods are to be submitted as a docker file. We will provide a docker template and submission guidelines by late August 2024.
2. Can I submit to only a specific subchallenge?
No, submitted methods will be evaluated across each of the 3 subchallenges.
2. Can I submit different methods to different subchallenges?
No, a single method will be evaluated against different criteria in each of the subchallenges.
--- Publication ---
1. Will my challenge submission be published?
We plan a joint publication of the CVS Challenge which will include the submitted challenge models and results. More information will be provided on this as time goes on.
2. When can a participant publish independent research on this dataset?
Participants are allowed to publish their results separately only after the publication of a joint challenge paper.
3. When will the joint results be published?
This is tentatively planned to happen before the end of 2025.