Stitching large dataset issue

[ctrueden]

From an ImageJ mailing list thread:

Doing some stitching of a "large" dataset I encountered a « bug ».

Stitching parameters

• Issue : Weird pixel attribution
• DATASET :
  • Channel : 1
  • Each Z-stack: 2048x2048x92 (around 750Mb / stack)
  • Total Positions : 112 (16x7)
  • Eachtiled-slice is around 700Mb,(total around 85 Go)
• Downscaled dataset
  • Channel : 1
  • Each Z stack: 512x512x23 (around 11Mb / stack)
  • Total Positions : 112 (16x7)
  • Each tiled-slice is around 30Mb, (total around 1.25 Go)
• Computer :
  • CPU IntelXeonx64, 16 Cores, 2.6Ghz,RAM 128Gb
  • GPU : Quadro6000

If the stitching of the original dataset is done with computing the overlap. We observed a weird pixel attribution. a bug ?

If the stitching of the original dataset is done without computing the overlap. We don’t observed anymore the bug.

Using a downscaled version of the data set, we don’t observed a bug if the stitching is done with or without computing overlap