Hi, are there any restrictions using pretrained model for further fine-tuning? Could pretrained model have been trained by using data from medical domain and similar task, i.e., segmentation ?
Created by kkirtac Dear ZhenliangNi,
please refer to our reply given in the other thread that you created.
Kind regards,
the ROBUST-MIS organizers Dear sir,
Hello, we want to participate in the instruments segmentation task . But we didn't find the label in the dataset. The label of images in each package named "10s_video.zip" can not be found.Do we need to manually annotate the labels?
I hope you can give relevant tips. Thank you. Sorry, we have to refine what we said earlier: You are allowed to use additional training data that is **publicly **available and was released **outside the field of medicine**. All datasets that were used have to be listed in a method description that has to be submitted along with the docker image for the final submission. You are allow to use the training data from this challenge **and** use additional training data from anywhere (except from previous endovis challenges or data from the medical field).
Examples for what is ok:
* use training data from this challenge
* use training data from this challenge **and **image net (outside the medical field)
* use imagenet only
Examples what is not ok:
* use training data from previous endovis challenges
* use training data from this challenge and data from previous endovis challenges
* (any kind of other data inside the medical field) **and** data from previous endovis challenges Thank you. If I understand correctly, we are **not** allowed to use **both** training samples (images and annotations) **and** publicly available pre-trained models which used data from previous Endovis challenges. Is that correct? Dear kkirtac,
the only restriction is not to use data from previous EndoVis challenges. Otherwise, it's fine and you are allowed to use data from the medical domain for training.
Kind regards,
the ROBUST-MIS organizers