Most requirements can be installed via the requirements.txt file using

pip install -r requirements.txt

However there were some modifications made to the BioPython Library. The modified _utils.py file is stored in the top level directory of this repository and should be used to replace <your installed packages>\Bio\Phylo\_utils.py in case you are working with venv then it would located at venv\Lib\site-packages\Bio\Phylo\_utils.py.

Below is a list of all config options and a brief explanation. The config file is config.json. It is recommended to first take a look at guide/guide.pdf

The directory where the frames of your video are located

Threshold to create a mask for inpainting (see guide/guide.pdf for more details)

Regulates how much of the local environment is considered when inpainting

After inpainting a gaussian blur is applied. The region_increase_for_blurring option will determine by how much the bounding box - region is enlarged. Blurring is then performed inside this enlarged region.

The amount of times blurring is performed

After the created template is subtracted from the current image. Thresholding is performed to obtain the Ecoli on the current picture.

After thresholding a first detection is performed and bounding rectangles are created. All boxes whose height is less than or equal to height_cutoff or whose width is less than or equal to width_cutoff are filtered out.

Ecoli can only appear or disappear from the frame if they are close to the border. This option determines the maximum distance in pixels

If there are two boxes at time $t+1$ whose Intersection over Area score for a box at time $t$ is above this threshold, duplication is assumed

Sets the minimum Intersection over Union score for a box at time $t$ to be mapped to a box at time $t+1$

If we "lose" a cell or merging happened we try to find the cell again by performing sliding window search. The search region will be it's old bounding box increased by a certain amount in height and width specified by these options.

In line 492 in _utils.py you might want to adjust the second value according to the duration of video. Depending on the duration maybe also other offsets need to be changed. It might be necessary to modify line 515 and line 435 to adjust the offsets for the labels

All those parameters may potentially be required to be finetuned when working with a new video.

When labelImg opens you first need to press Open Dir and select the folder with your images (usually ./images) and then press Change Save Dir to select the directory where the label files are being stored. This directory should also contain a file called classes.txt which contains all the different classes. For now it would just look like this

Ecoli

If you have everthing set up you can start tracking with

python track.py

and if you just want to use the code for generating labeled training data you can run

python automatic_labeling.py

Once your tracking has finished you can export it with

python export.py

If you have any questions regarding the source code just contact me on Discord: elson1608.