Known issue: Difference between revisions
No edit summary |
No edit summary |
||
Line 1: | Line 1: | ||
== Release 1.3 == | |||
No known issues yet... | |||
== Release 1.2 == | == Release 1.2 == | ||
Revision as of 11:36, 18 June 2014
Release 1.3
No known issues yet...
Release 1.2
In some cases refinement runs crash after 1 or 2 iterations and all intensity scales in the data_model_groups table become NaN. The actual cause lies elsewhere, but a good solution is changing line 1890 in src/ml_optimiser.cpp:
mymodel.scale_correction[igroup] = 0.;
should be:
mymodel.scale_correction[igroup] = 1.;
This bug does not affect runs that do not crash.
Release 1.1
Reconstruction sometimes gives strong stripy artifacts
This is a nasty bug in the reconstruction algorithm, which mostly seems to affect reconstructions from many particles, and only the 3D auto-refine option. The effects on the reconstruction can be disastrous: your molecule may disappear completely and you may end up with a stripy pattern instead. Note that if this happens in only one of the two half-reconstructions, it may go by unnoticed (if you don't inspect both half-reconstruction at each iteration) and may just appear as a sudden drop in resolution during the 3D auto-refine optimization. Somehow, this bug is system-dependent and sometimes hard to reproduce.
However, it seems the bug can be circumvented by changing line 658 in src/backprojector.cpp
from:
double myfsc = XMIPP_MAX(0., DIRECT_A1D_ELEM(fsc, i));
to:
double myfsc = XMIPP_MAX(0.001, DIRECT_A1D_ELEM(fsc, i));
Then recompile relion using:
make -j 4 && make install
Thanks to Dieter Blaas (Max F. Perutz Laboratories, Vienna) for reporting this bug. It will be fixed in the next release.