Research to track segmented 3d images of protein clusters.
2019 CVPR. Multi-Object Portion Tracking in 4D Fluorescence Microscopy Imagery with Deep Feature Maps. Paper link: http://openaccess.thecvf.com/content_CVPRW_2019/papers/CVMI/Jiao_Multi-Object_Portion_Tracking_in_4D_Fluorescence_Microscopy_Imagery_With_Deep_CVPRW_2019_paper.pdf
| Raw Sprites Dataset | Labeled Sprites Dataset |
|---|---|
 |
 |
Not a perfect demostration as annotated datasets with many to one or one to many tracking are non-existent. Full video in ./data
| Protein Dataset 2D Distribution | Protein Dataset Clustering |
|---|---|
 |
 |
In the left image we see the 2D mapping of the feature embeddings. In the right image we see each cluster labled by DBSCAN. The cluster classes have sizes:
| Class Labels | Cluster Sizes |
|---|---|
| 0 | 16773 |
| 1 | 994 |
| 2 | 2472 |
| 3 | 281 |
Total feature embeddings: 10260
To set up python3 virtual env.
/cell_tracking$ python3 -m venv env
/cell_tracking$ source env/bin/activate
/cell_tracking$ pip3 install -r requirements.txt
This requires the original segmentation results data, from the 2019 CVPR Multi-Object Portion Tracking in 4D Fluorescence Microscopy Imagery with Deep Feature Maps paper.
/cell_tracking$ python3 data.py --h
Running this will generate a numpy dataset from the raw *.nii files. Will save in /data folder. This is used by the tracker.
/cell_tracking/src/python$ python3 protein_tracker.py --task train
/cell_tracking/src/python$ python3 protein_tracker.py --task predict
This will run tracking on the dataset and store the results in data/ as json files.
labled_tracks.pickle: Contains a python dict that holds the tracking results
tracks_frame.json: Json file where the key is the frame number and the value is a list of every track in that frame.
tracks_pretty.json: Json file where the key is the cluster id and the value is a list of all the frames that it appears in.
Run the Matlab script graph_results.m to label the 3D figs with the tracked ID's. This uses the json tracking result files.
Copy labled .fig results to cell_tracking_research/data/labled_frames
/cell_tracking_research$ cp ./data/raw_data/Segmentation_and_result/*/*_tracked.fig ./data/labled_frames/
Run demo.m or open demo.fig in /data to see tracking results, slide the slider the change frames.