Offset and resolution in groudtruth data

[heliy]

Hi, thanks for sharing data! I meet problems when I try to map the groudtruth (GT) crop into the raw volume. I have checked suggestions from issue #1 and issue #4 but still failed.

First question is that I cannot find the location of GT crops in raw volume. I take the jrc_sum159-1/crop20 crop to present messages I got from your documentations and issues.

• raw_attrs is the loaded metadatas from volumes/raw/s0
• gt_attrs is the loaded metadatas from volumes/groudtruth/0003/crop20/labels/all

import numpy as np

raw_scale = np.array(raw_attrs['transform']['scale']) # [4.56, 4, 4] (nm), resolution of raw, in zyx order
# GT-related metadatas are in xyz order so I use `[::-1]` to reverse them
gt_scale = np.array(gt_attrs['resolution'])[::-1]          # [2, 2, 2], resolution of GT
gt_size = np.array(gt_attrs['dimensions'])[::-1]          # [170, 400, 400], size of GT
gt_offset = np.array(gt_attrs['offset'])[::-1]              # [22004,  2980, 18820] (nm), location of GT

# gt_offset/raw_scale yields the voxel-level offset in raw volue
voxel_offset= gt_offset/raw_scale          # [4825,  745, 4705]

Next, I crop a [1x200x200] image from raw at the voxel-level offset and resize GT[0] (the 1st z layer) into [200, 200] to check EM image and annotations. Unfortunately, it seems the offset is error.

Another question is about the "z" resolution. In GT crop it is 2nm while in raw volume it is 4.56nm (or 5.24nm/... in other datasets). Suggestions in previous issues use [::2] downsample in z-axis, not matching metadatas. I'm confused about it as I found the volumes/raw says it is 4nm, and the page in openorganelle says it is 4.56nm.

Should I use the 2/4.56 ratio to calculate the size of GT crop in raw? or use the [::2] downsampling (1/2 ratio) and discard differences among datasets? Which following is correct?

# solution 1: use the 2/4.56 ratio
shape = gt_size*gt_scale/raw_scale # [75, 200, 200]

# solution 2: use the 1/2 ratio
shape = [int(gt_size[0]/2)]+list((gt_size*gt_scale/raw_scale)[1:]) #[85, 200, 200]

Thanks again for your amazing datas.

Best wishes!

Liuyuan He.