Hi,
https://gitlab.com/nct_tso_public/challenges/miccai2024/FedSurg24/-/tree/main?ref_type=heads
According to the above link, the data folder is expected to have the following structure:
- data
|--- train
| |--- client1
| | |--- data.csv (image_id, class)
| | ?--- frames
| | |--- frame1.jpg
| | |--- frame2.jpg
| | ?--- ...
| |--- client2
| | |--- data.csv
| | ?--- frames
| | |--- frame1.jpg
| | |--- frame2.jpg
| | ?--- ...
| ?--- ...
|--- test
| |--- client1
| | ?--- frames
| | |--- frame1.jpg
| | |--- frame2.jpg
| | ?--- ...
| |--- client2
| | ?--- frames
| | |--- frame1.jpg
| | |--- frame2.jpg
| | ?--- ...
| ?--- ...
However, when I see trainset at https://gitlab.com/nct_tso_public/challenges/miccai2024/FedSurg24/-/tree/main/data_path/trainset?ref_type=heads
it has a different structure:
it is divided into Center01, etc. and then each center has .csv file and corresponding images.
Both formats seem to be very different. Can you please help in clearing the problem here. Thank you.
What is the final decided format for training and testing?
Also, if the first one is the format for training, should we assume the center info is lost, each video is a client and can be from any center. Also, data.csv corresponds to that particular client which means it only has 1 row?
Please clarify.
Thank you,
Santhi Kolamuri
Created by Santhi Raj Kolamuri SRajKolamuri Thank you.
Can you please comment on the data format for Center_04? Would it be the same? Please clarify.
Thanks,
Santhi Hi Santhi,
sorry for the confusion. The Git repository was created before we had the challenge data.
Therefore, please use following data structure:
.
??? center01
? ??? Train_Private
? ? ??? 1 (folder contains the 200 images)
? ? ??? 12
...
? ? ??? train_private.csv
? ??? Train_Public
? ??? 14
...
? ??? train_public.csv
??? center02
...
? ??? Train_Private
? ? ??? 10
...
? ? ??? 9
? ? ??? train_private.csv
? ??? Train_Public
? ??? 16
...
? ??? train_public.csv
??? center03
? ??? Train_Private
? ? ??? 1
..
? ? ??? train_private.csv
? ??? Train_Public
? ??? 106
...
? ???train_public.csv
The testing format will be the same.
.
??? center01
? ??? Test
? ???1
| ???images (1 -200)
...
??? center02
...
??? center03
? ??? Test
? ???1
...
You decided the quantity of images you would like to take for inference, e.g., only the center frame, n frames, or all 200 provided frames.
During testing we would need an output which looks similar to one of the csv.
Example:
center, surgery, label
center01, 1, 4
...
For Task 1 I would use the global model and for Task 2 the adapted model from each center.
I hope that clarifies your questions.
BR Max