Datasets:
Clarification on tile count, label range, and 512px patch availability
Hi 👋
First of all, thank you for releasing PANDA-PLUS-Bench. It's nice that you open-source higher quality labels than what PANDA originally shipped (known to be noisy). I’m very much interested in using it for my research.
While preparing the dataset, I ran into a few discrepancies and wanted to clarify them:
1) Number of 224px tiles
The paper and dataset description mention 2,770 224px tiles, but the dataset currently available appears to contain 3,872 tiles. Does the released dataset include additional samples beyond those reported in the paper?
2) Label range discrepancy
The paper/dataset description states that labels range from:
0= benign1–3for Gleason 3 / 4 / 5 patterns
However, the dataset labels range only from 0 to 2. Could you clarify why?
3) 512px patches
The paper mentions 512px patches. Do you plan to release these as well? If so, is there a tentative timeline?
4) Full annotation masks
I was also wondering about the availability of the full annotation masks. Do you plan to release the full-resolution annotation masks? If not, would it be possible to obtain them upon request for research purposes?
Thanks again for making this benchmark publicly available!
clement