Skip to content

Run Bigstream2 in Pipeline #48

@anaste123

Description

@anaste123

Bug report

Hi, I was trying to do registration of our large image data using Bigsteam2 in EASI-FISH Pipeline, I changed most parameters exposed, including:

    --bigstream_global_steps
    --bigstream_local_steps
    --global_iterations
    --local_iterations
    --local_ransac_cc_radius
    --local_ransac_nspots
    --local_ransac_match_threshold
    --local_ransac_diagonal_constraint
    --aff_scale
    --def_scale
    --global_ransac_fix_spot_winsorize_limits
    --global_ransac_mov_spot_winsorize_limits
    --local_ransac_fix_spot_winsorize_limits
    --local_ransac_mov_spot_winsorize_limits

I also tried the parameters you recommended before:

屏幕截图 2024-01-21 152127
fig2

The same image can get good registration results using Bigstream2 Python script, but can not get good results using Bigstream2 in EASI-FISH Pipeline. From my experience, I think shrink_factors, smooth_sigmas from affine_kwargs/deform_kwargs have great influence on the registration results.

Could you please expose those parameters in Registration step? Is there any other parameters exposed that I can use to improve the registration results? Thanks in advance!

Description of the problem

Log file(s)

Environment

  • EASI-FISH Pipeline version: [dev]
  • Nextflow version: [23.04.1]
  • Container runtime: [Singularity]
  • Platform: [Slurm]
  • Operating system: [CentOS 7.6]

Additional context

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions