vEMstitch

Here is an official code for vEMstitch, an algorithm for fully automatic image stitching of volume electron microscopy. We provide both python versio and C++ version.

Requirements

Python:

We have provided the "environment.yaml" for Anaconda.

conda env create -f environment.yaml 

C++:

OpenCV, cmake

Usage

Python:

In the "source" directory, there is a main.py file.

Usage: python source/main.py --input_path [data dir] --store_path [result dir] [(options)]

---options---
--pattern [int (N)]: The N*N image stitching
--refine: feature re-extraction or not

C++:

cd vEMstitch_c++
mkdir build
cd build
cmake ..
make

cd ../../
vEMstitch_c++/bin/vEMstitching [data_dir] [store_dir] [pattern] [overlapping_rate] [log_dir] [refine_flag]

We strongly recommend that user first use the tool without "--refine" to fast and robustly stitch tiles. Then, if there are misalignments in some results, users can specifically re-run the tool with "--refine" to acquire seamless images.

Testing example

We also provide a testing example in the "test" directory.

Without refinement

Python:

python source/main.py --input_path test --store_path test_res --pattern 3

C++:

vEMstitch_c++/vEMstitching test test_res 3 0.1 log.txt false

With refinement

Python:

python source/main.py --input_path test --store_path test_res --pattern 3 --refine

C++:

vEMstitch_c++/vEMstitching test test_res 3 0.1 log.txt true

related_data

simulated_data

for illustration, some simulation results (three examples of total 100 ones)

raw_data: raw single image

simulation1: translation only

simulation2: translation + rotation

simulation3: rigid + local distortion

simulation_noise: different level of noise

simulation_deformation: different level of deformation

C_: section image

C*_res: result of vEMstitch

C_stitching_row: row result of vEMstitch

fiji_restore: result of Fiji

mist_restore: result of MIST

trakem2_restore: result of TrakEM2

real_test

we have provided the raw sections and stitched results of compared methods.

The all real mussel images used in the paper are available at https://pan.quark.cn/s/f097018cdf7b.

Licenses

[]

<!-- The data is licensed under [Creative Commons Attribution-ShareAlike 4.0 International License](http://creativecommons.org/licenses/by-sa/4.0). -->

We assign Licenses to the code and data separately. The code matched by the following patterns are licensed under GPL-3.0:

• *.py
• *.yaml
• *.cpp
• *.h
• *CMakeLists.txt

The simulation data based on CREMI dataset(https://cremi.org/data) and real mussel images are available under CC0 1.0, including:

• *.bmp