For example, we have detected that the following file ([idx_frame,] idx_slice) has all 0 values. They will cause the network to collapse.
```
TaskR1 T1w/ValidationSet/UnderSample_TaskR1/Center005/GE_15T_voyager/P007/T1w_kus_ktRadial24.mat (10)
TaskR1 T1w/ValidationSet/UnderSample_TaskR1/Center005/GE_15T_voyager/P009/T1w_kus_Uniform24.mat (11)
TaskR1 T1w/ValidationSet/UnderSample_TaskR1/Center005/GE_15T_voyager/P012/T1w_kus_Uniform8.mat (9)
TaskR1 T1w/ValidationSet/UnderSample_TaskR1/Center005/GE_15T_voyager/P023/T1w_kus_ktRadial8.mat (8)
TaskR1 T1w/ValidationSet/UnderSample_TaskR1/Center005/GE_15T_voyager/P023/T1w_kus_ktRadial8.mat (9)
TaskR1 T2w/ValidationSet/UnderSample_TaskR1/Center005/GE_15T_voyager/P007/T2w_kus_ktGaussian16.mat (10)
TaskR1 T2w/ValidationSet/UnderSample_TaskR1/Center005/GE_15T_voyager/P012/T2w_kus_ktRadial16.mat (9)
```
Therefore, as of now, the evaluation results submitted to the synapse verification machine seem to be unreliable.
We wonder if the test set data currently under ranking contains the same problem?
Created by anonymously hachimen We have already checked the issue you mentioned, and it is true that some frames contain all-zero data. However, we believe that the all-zero data will not affect the overall reconstruction score. Moreover, such data is extremely rare and only appears in the last one or two frames. If this affects your model, we suggest skipping the processing of those frames. However, please ensure that you merge the data correctly during the final submission to maintain the proper data dimensions and guarantee valid scoring.
During the testing phase, we have additionally checked the situation you mentioned and will try to avoid all-zero data as much as possible. However, there is still a possibility that a very small number of frames may contain all-zero data. We appreciate your understanding.
Drop files to upload
Validation set contains corrupted data page is loading…