In that case the error seems to stem from a bug in how QSIPrep was calculating the extent window for dwidenoise. With a run of 25 volumes, it was setting the extent window to 1x1x1, which produces a map of all zeros. With a run of 27 volumes, the extent window is 3x3x3, so there won’t be an error. We just merged a fix yesterday and will hopefully make a release later today.
1.0.2 is now up on DockerHub, so can you try updating to that and re-running?
Thank you for your prompt fix! I have already downloaded and used the latest 1.0.2 version from DockerHub, and I can confirm that the 25-direction dataset can now be processed successfully.
However, I encountered a new issue: only the first subject runs successfully. Starting from the second subject, every dataset fails at the synseg step. The same data did not have this problem with the previous version. My docker has 40 CPUs and 251.5 GiB of total memory.