Hello,
1. I would like to confirm that the key frame (center 2, folder 44) is still 100? As the total number of frames in this folder is 199.
2. Does the final test evaluation include 200 frames as input to the model, or only the key frame?
3. Are segmentation masks allowed to be created?
4. Is there any issue with using publicly (or privately) pre-trained weights?
Thanks!
Created by Julia Alekseenko ajulyav @kirchnermax
Thank you for the details!
Could you please confirm the number frame (i.e. 0001, 0002...) for IMG_3853.png updated to Center 2, Folder 44?
Thank you once again! 2.)
We decided that per video one inference is necessary. You can decide if you just use the Center frame, n frames, or all 200 frames. Hi
1.) We checked the challenge data and one file wasn?t uploaded. We will try to upload it. So there will be 200 images again.
2.) TBD
3.) You can create segmentation masks as intermediate result. The model is evaluated on just the inflammation level.
4.) In the challenge design is following written:
Define the policy on the usage of training data. Only the provided training data and publicly available data, including open, pre-trained networks, may be used. Due to the federated learning nature of the challenge, most training data will be kept private. Some examples will be provided to develop and select the methods used in the federated learning part of the challenge. This provided data may not be used to pretrain the models used in the federated learning part of the challenge