processing notes

[meghalithic]

• process non-50 mag separate
• always have same padding around scale
  • run rotated pictures separately

[meghalithic]

I separated images into folders based on the species. I also am rotating images so that all zooids are more top-bottom oriented rather then left-right.

I am currently going through the images select a subset of these for analyses

• removing images with lots of secondary calcification
• removing images not straight on
• removing images of colonies that are not flat
• removing colonies that are broken
• removing colonies that are dirty
• removing images with ancestrula
• selecting colonies with good definition of zooids

If too few colonies, can go through the omitted images and reassess.

[meghalithic]

post-processing:

• remove autozooids without an orifice or avicularia
• remove orifices and avicularia without an autozooid
• remove autozooids (and their parts) that overlap a lot
• maybe do an outlier test and remove really large and really small autozooids (and their parts)

[meghalithic]

pre-processing notes:

• re-add images of competition

• sensitivity of deep bryo

  • really sensitive to calcification
  • absolutely remove broken and really dirty zooids/colonies and those with too much calcification
  • maybe keep some of the colonies at an angle to retain zooids that are ok

• how strict is autofiltering?

  • too strict
  • decided to only use padding, 5 pixels per side and 125 for scale bar

• if have too many boxes to deselect

  • make note of the image and which boxes to remove or keep

• selecting more than 16 boxes to exclude causes it to crash

• when using DeepBryo, exclude 'bad' masks for avicularia (i.e., broken) and orifices

• when using DeepBryo will want to pay special attention to the ovicell as that will be hard to discern how 'good' the mask is post-processing

• will want to look at overlapping by mask and not by bounding box as the bounding box can overlap with adjacent orifices and avicularia

• remove apesia and spiramen

• does it matter how far away it is?

[lilysandstrom]

• keeping overhanging ovicells
• the slant of the orifice - 30 or 50%
• removing masks affected by dirt
• calcification that greatly enlarges the zooid or gives the orifice a "hat"
• removed things where the person operating could not tell if it was that feature
• keep the masks with holes
• broken avicularia where it expands the area beyond what is natural
• keep the avicularia where the size is different due to possibly taphonmy but they still look like avicularia
• keep as much as possible
• if the image has too many issues, make note on the GitHub issues, and move on
• removing one row that's too close to the ancestrula
• keep very small zooids that are not ancestrula as long as they have classic microporella shape
• keeping the zooids on the edge of the colony as long as area is okay and avicualria are either not detected or removed
• obvious slanting of the zooids removed
• keeping broken zooids if the mask is detecting it fine
• removing ones with cracks through the zooids in the colony, but keeping ones with cracks that don't affect the zooids
• removing the zooids that are a different species
• removing edge cases

[lilysandstrom]

Organization: Keeping all images on the desktop -i ~/Desktop/microporella-subset-jpg and then call each species subfolder with appropriate padding All outputs will go into the Microporella Data folder on GitHub -o ~/Documents/GitHub/bryozoa/microporella/Data CLI-outputs will go into at folder in GitHub called CLI outputs with each species as a subfolder Manual outputs will go into a folder in GitHub with each species as a subfolder

[lilysandstrom]

seems to not be a big difference when we're selective compared to the client spend our time on the following:

1. orienting the images
2. manually focus on images with other species (make list in a spread sheet to know to go back and make list of ids to remove)
3. run everything through the cli
4. run the images of that have other species manually
5. then if time focus on images where autozooids are really "bad" (i.e., over estimated due to weird shape or calcification)

[lilysandstrom]

Post-processing:

• Remove orifice/avicularia (or whole autozooid) if the detected feature covers more than 50% of the autozooid
• Remove if zooid has more than one orifice
• Remove autozooids that overlap too much with other autozooids
• Remove orifices that overlap with other orifices, ovicells that overlap with other ovicells, avicularia that overlap with other avicularia, and ascopores that overlap with other ascopores.
• Remove ascopores that are size-outliers, eg. bigger than avicularia