fidoptDeepCLEM, fluorescence and SEM with a known deformation
A fluorescence and a scanning electron microscopy image of one C. elegans section, aligned by hand by the provider, the SEM image deformed by a seeded smooth warp whose recovery is measured.
4 test pairsC. elegans sectionssame sectionfixed FLUORESCENCE, moving EM2D, distances in pxdelivered aligned
Read before comparing (6)
- Truth. Synthetic: one hundred seeded points on the fluorescence foreground, at least 18 pixels from the border, paired through the inverse of the added B-spline warp. They measure recovery of that warp relative to the provider's manual alignment, not independent anatomical correspondences; the alignment's own error is not removed.
- Points. The fluorescence image is a few bright spots on black, and the points sit on them: on s6_ds, four clusters, not the whole field.
- Baseline. With no registration the warp leaves 5.4 to 9.4 pixels on the four test tiles; ANTs SyN ends above that start on s7_ds and s9_ds.
- Units. The TIFFs have no usable pixel size, so spacing is one and distances are pixels; the warp's amplitude_mm and spacing_mm, 16 and 64, are pixels too.
- Images. Some SEM TIFFs hold two identical planes: prepare.py keeps one after checking they are equal and rejects a TIFF whose planes differ. The images are not cropped; the 18-pixel border keeps both points of every pair inside the field under the 16-pixel warp.
- Cases. The four validation pairs are not multiplied into more specimens by crops or random warps: with so little specimen information the dataset is a synthetic track of its own.
No method has a result on every pair of this dataset yet.
Pair by pair
Click a score to look at that result: the fixed image, and over it the moving image after registration, with the annotations. Underlined scores have their image; for the others you see the pair without the result. A green cell gained against no registration on that pair, a red one lost.
About the data
The first test caseClick to look at every case.
- Test pairs
- The provider's four validation pairs are the test cases. Ten development pairs are drawn from the A1/A2 training series; the s-series training tiles are left out, as they are named like the validation sections s6 to s9. Animal identities are not given, so animal independence is not established.
- Truth
- landmarks a known deformation One hundred seeded fluorescence-foreground points at least 18 pixels from the border, paired through the inverse of the added known B-spline deformation. These measure recovery relative to the released manual alignment, not independently measured anatomical correspondences.
- Licence
- CC BY 4.0
10 pairs for tuning, never tested
A2S5_U, A2S6_U, A1S10_U, A2S7_U, A1S8_U, A1S6_U, A2S3_U, A2S12_U, A1S1_U, A2S2_U