Data format

Hi,

I have a question regarding the general video data, as I noticed that in the training data the used devices are defined at the bottom of each frame and oftentimes only two devices are visible in one 30s clip but according to the GT label three devices should be present. Is this intentional? Further, I wonder if the test data will have the same descriptive caption in each frame at the bottom.

Hi TS_UKE,

Thank you for your question.

That's a correct observation - the tools visible within the view may not be 3 at all times even if the label says that. This is a noise within labels that gets introduced with automatically extracting tool information from robot system data. The tool within the label is installed but the surgeon might have moved that tool away if it is not being used.

The test data will not have the user interface including the descriptive caption underneath and will be cropped out.

Please feel free to reach out again with follow-up or any other question.

Best, SurgToolLoc 2022 Organizing Committee

Hi Aneeqzia_isi,

thank you for your fast reply. I have indeed some follow-up questions/remarks.

1.) Can you provide an example for the test data (i.e., clip + GT-frame labels + bbox)? 2.) At which point a device that moves into a frame is considered as visible? 3.) In the data description clips are described as being captured with 30 fps, however, loading all frames of one 30s clip using opencv returns 1800 frames (indicating a capture rate of 60fps).

Thanks again and best regards TS

Hi TS_UKE,

Apologies for the delay in responding back. Following are responses to your questions:

1. We plan to upload submission instructions which will include details regarding the format of output files etc. However, based on questions from teams, we are trying to upload a few ground truth bbox labels for some videos within the training set. We will announce it and let everyone know once its uploaded.
2. It depends on specific tool type as well as mentioned in data description. The tool clevis is the main part of interest within this dataset (examples shown at https://surgtoolloc.grand-challenge.org/data/). If the tool clevis is not visible and just a small tool tip is within view, that will not be considered visible within our test set annotations.
3. Thank you for pointing this out. The videos are in fact at 60fps - the description on the website has been corrected.

Hope this answers your questions. Please do not hesitate to reach out with follow-up or new questions.

Best, SurgToolLoc 2022 Organizing Committee

Hi anniqzia_isi,

I would like to follow-up on your last message where it was mentiond " we are trying to upload a few ground truth bbox labels for some videos within the training set". Are the bounding boxes examples uploaded somewhere already?

Kind regards,