search

MIRACLE-Center / CTPelvic1K

Resources of the paper “Deep Learning to Segment Pelvic Bones: Large-scale CT Datasets and Baseline Models”.
129 stars 35 forks source link

Discrepancy in COLONOG Dataset Mask Count: Stated 731 vs. Found 714 #28

Open Ziyan-Huang opened 1 year ago

Ziyan-Huang commented 1 year ago

Thank you for your great work in providing such a large-scale annotated dataset for the pelvis.

While working with the datasets provided, I noticed a discrepancy in the count for the COLONOG dataset. According to both the repository documentation and the associated paper, the count for this dataset is stated as 731. However, after downloading and inspecting the data, I found that there are only 714 masks. image

image

image

Could you please check it for me? I'd really appreciate it.

Best Ziyan Huang

e-hicks-ntv commented 1 year ago

Hi Ziyan,

Apologies to ask a question to your question, but how did you access the COLONOG dataset? I can find the masks but not the images. There is the TCIA dataset but that's 400+ GB which I'm not sure I could download in one go. Have you downloaded the CTs from a different source?

Any advice on working with this dataset would be greatly appreciated.

Thanks, Elliot

Ziyan-Huang commented 1 year ago

Hi Elliot,

I haven't accessed the original data directly from the official website either. However, I found that CTSpine1K provides the raw data for this dataset. You might need to contact CTSpine1K via email for access. That said, I noticed some discrepancies between the mask and image sizes. The CTPelvic dataset seems to be challenging to use with annotations provided for other public datasets.

However, I think training and validating on the provided CLINIC data (dataset6) should suffice. Some papers have done it this way.

Best, Ziyan

e-hicks-ntv commented 12 months ago

Hi Ziyan,

Thank you for getting back to me so quickly, I'll do that!

Thanks, Elliot