ryhattori / patchwarp Goto Github PK

First, thank you again for this great toolbox and the last fix for two-channel data.
Working with your toolbox, I found another potential bug. When using two-channel data and setting ops.n_ch = 2, ops.run_rigid_mc = 1, and ops.run_affine_wc = 1, when calling patchwarp(ops), the resulting images contain only a quarter of the original.
The expected behavior would be half of the images.
The issue is related to the parameter n_ch being passed to both correction functions.
My quick and dirty fix is to hardcode ops.n_ch = 1 in the patchwarp_affine.m file. This should make the most sense for most applications.
Sincerely

• Kilian Rolle

Hi, thanks again for putting up the code for PatchWarp. Using most recent PatchWarp main branch on Matlab 2020a. PatchWarp encounters an error within patchwarp_affine when attempting to call variable warp_success with certain movies.

It appears this is because warp_success is defined as a scalar/binary output in ecc_patchwarp, see

And yet in patchwarp_affine it is called with (warp_success(i1,i2,i3) == 0), or as a 3D matrix leading to index issues. This error usually is not encountered as the short-circuit || leads to that expression not being called since it is second.

This could potentially be fixed if warp_success in the below line is changed to warp_success(i1,i2,i3).

Running the rigid motion correction works fine. I'm using the demo file as a template. After finishing this part and starting the warp correction, I receive the following warnings and errors:

Loading downsampled_perstack file...

Warning: Total frames are not a multiple of n_ch.

In read_tiff (line 62)

In patchwarp_affine (line 55)

In patchwarp (line 150)

In patchwarp_demo (line 123)

Estimating affine transformation matrices...

n_split4warpinit is too large
n_split4warpinit was changed to 4
Applying affine transformations...
Error using patchwarp_affine
warp number does not match file number

Error in patchwarp (line 150)
patchwarp_affine(save_path_prewarp, save_path_postwarp, ops.n_ch, ops.align_ch, ops.affine_norm_radius, ops.warp_template_tiffstack_num, ops.warp_movave_tiffstack_num, ops.warp_blocksize, ops.warp_overlap_pix_frac, ops.n_split4warpinit, ops.edge_remove_pix, ops.affinematrix_abssum_threshold, ops.affinematrix_abssum_jump_threshold, ops.affinematrix_rho_threshold, ops.affinematrix_medfilt_tiffstack_num, ops.transform, ops.warp_pyramid_levels, ops.warp_pyramid_iterations, ops.downsample_frame_num, ops.worker_num, ops.network_temp_copy);

Error in patchwarp_demo (line 123)
patchwarp(ops);

I would appreciate an idea of the issue and a potential solution.

It seems like this doesn't support Big Tiffs. When I input a bigtiff, it only spits out a 4 GB tiff file for rigid correction, and I get an error with the nonrigid.
Do you forsee any problems if I switch the tiff read/write support utilities to those from my github repo, supporting bigtiffs? https://github.com/jmstujenske/Matlab_FastTiffReadWrite It also have a TiffViewer utility based on memory mapping that can load the tiffs quickly for checking the motion correction quality, which may be helpful.

Hi,
I am excited to try out this new motion correction algorithm. Just wanted to report a small issue, namely that java.io.File fails if paths are specified as character vectors and not strings. I got this error from the fastdir function:

No constructor 'java.io.File' with matching signature found.

Error in fastdir (line 11)
    jif = java.io.File(dir_name);