search

JunMa11 / AbdomenCT-1K

The official repository of "AbdomenCT-1K: Is Abdominal Organ Segmentation A Solved Problem?"
https://ieeexplore.ieee.org/document/9497733
Apache License 2.0
227 stars 35 forks source link

Annotation data does not correspond to image data #6

Closed shchojj closed 3 years ago

shchojj commented 3 years ago

First of all, thank you for the AbdomenCT-1K dataset, which really solved the problem of partition generalization I encountered.However, when I downloaded the data, I found that the Mask data did not correspond to the Image data, for example, the Image data was 1~1000, but the maximum file of the Mask data was 1062, and the total number of mask was 1000. How to correspond to these data?

shchojj commented 3 years ago

In addition, there are some problems with some data, such as data "00433". When I read data through SimpleItk, I said "No Orthonormal Definition found!"

JunMa11 commented 3 years ago

the Image data was 1~1000,

We invite some guys to download the dataset but we can not reproduce your problem. The number of image cases is 1062.

When I read data through SimpleItk, I said "No Orthonormal Definition found!"

It is interesting. Basically, we usually use itk-snap and nibabel to handle the data. Both of them work well with this case.

shchojj commented 3 years ago

Thank you for your reply. Sorry for that, I didn't unzip the "Case_1000-.7z" file. However, there are only 1000 Mask data, which does not correspond to the number of image data (1-1063). However, a cursory reading of the data set still shows an error "No Orthonormal Definition found!" Data such as the following: ['00433','00405','00051','00702','00232','00423','00432','00401','00240','00244','00217','00403','00438','00687','00213','00549','00251','00681','00205','00221','00756','00448','00011','00049','00712','00021','00057','00249','00238','00219']

JunMa11 commented 3 years ago

Please read the Readme. we keep the remained label hidden in the grandchallenge. You can obtain the evaluation results by submitting the results.

Does nibabel also have this problem?

shchojj commented 3 years ago

Thank you for your patience. Do you mean the "HiddenTestSet1063-1112.txt" file?50 cases (1063~1112) of Mark data are hidden, but there is no intersection with Image dataset 1~1062.Moreover, there are 63 marks difference between Mark data and Image data. Yep, "nibabel" reads fine, I prefer "Simpleitk" to read and write data, maybe I should switch to "nibabel" which is more compatible.

JunMa11 commented 3 years ago

Please check the readme file in the mask folder

To avoid the ground truth leakage in AbdomenCT-1K benchmarks, we made 1000 annotations publicly available. 
The hidden annotations can be accessed on the grand-challenge by submitting your segmentation results.
- Fully superivsed learning: https://abdomenct-1k-fully-supervised-learning.grand-challenge.org/
- Semi-superivsed learning: https://abdomenct-1k-semi-supervised-learning.grand-challenge.org/
- Weakly supervised learning: https://abdomenct-1k-weaklysupervisedlearning.grand-challenge.org/
- Continual learning: https://abdomenct-1k-continual-learning.grand-challenge.org/