fishfishson / co-tracker

Co-Tracker For 3D Cell Reconstruction From Monocular Microscope Video

Cell Pose Reconstruction

Usage

1. install related python libraries (e.g., pytorch, trackpy) following the instruction of CoTracker (see README-COTRACKER.md).
2. Include current workspace into your python env and go to the code dir.

export PYTHONPATH=`pwd`
cd custom

3. Arrage the data files into the following structure

$DATAROOT
├── Blebb
│   ├── Cell 12
│   ├── Cell 3
│   ├── Cell 5
│   ├── Cell 7
│   └── Cell 9
├── Cilio
│   ├── Cell 11
│   ├── Cell 17
│   ├── Cell 20
│   ├── Cell 5
│   └── Cell 8
├── DMSO
│   ├── Cell10
│   ├── Cell12
│   ├── Cell2
│   ├── Cell5
│   └── Cell8
└── YW

and for each cell dir, there exist

$DATAROOT/Blebb/Cell 12
├── After drug
│   ├── 12_Position 13 after _Crop001_RAW_ch00.pkl
│   ├── 12_Position 13 after _Crop001_RAW_ch00.tif
│   ├── MetaData
├── Before drug
│   ├── 12_Position 13 before_Crop001_RAW_ch00.pkl
│   ├── 12_Position 13 before_Crop001_RAW_ch00.tif
│   ├── 3d.pkl
│   ├── blender.pkl
│   ├── MetaData
│   ├── 
└── Cell to track.PNG

4. run codes to perform detection

# marker detetion
python3 detection.py --data_path $DATAROOT/DMSO/Cell2/After\ DMSO

5. (Optional) run segmentaiton After detection, you will get images.mp4. Use sem-and-track to segment it. Put the segmentation results into dir images_masks:

   ├── After drug
   │   ├── images_masks
   |   |    ├── 0000.png
   |   |    ├── 0001.png

6. run tracking and pose estimation

# tracking
python3 tracking.py --data_path $DATAROOT/DMSO/Cell2/After\ DMSO --start_time 0 --duration 200 --vis_threshold 0.1 --mode full
# pose estimation
python3 pose_estimation.py --data_path $DATAROOT/DMSO/Cell2/After\ DMSO --radius_mode sem

PS: Many thanks for yizh4ng's codes from https://github.com/yizh4ng/3D-reconstruction-from-nuclei-rotation.custom/