idtrackerai.ai

We use idtracker.ai to segment and track multiple individuals in the same arena

Once a recording has been completed, you can analyze with idtrackerai, like this:

Set analysis params in the GUI

  1. Open the imgstore with idtrackerai

open_idtrackerai

  1. Required input:

  2. Number of animals

  3. Area thresholds
  4. Intensity thresholds
  5. Session name (datetime)
  6. Apply ellipse ROI. Provide 5 interspaced points along the outer outline of the glass dome

-> Save parameters with name datetime.conf

  1. Open the datetime.conf file and remove

  2. Session name

  3. Video name
  4. Chunk number

Schedule chunk analysis

cd $ROOT_OF_VIDEOS_FOLDER

idtrackerai-loop --input `pwd`/./2022-03-11_22-41-00_ROI_0 --interval 0 5 --knowledge-transfer previous --environment idtrackerai4 --suffix "" --jobs 15

  • --jobs sets the maximum number of CPU cores to use. Please set it to a conservative value, depending on your total number of CPUs and the other processes which may be running in your PC.

  • --interval determines which chunks of your store will be analyzed. The second number is the first chunk which is NOT analyzed (i.e. as in range(0, 5), which means 0 1 2 3 4, and not 5)

  • --input must receive a path to a folder with an imgstore. The path should be split with /./, to indicate what is the root, and what is the imgstore folder itself. Everything after the dot will be hardcoded into the data, which still allows to move the imgstore around without the idtrackerai results breaking.

  • --environment should contain the name of a anaconda environment which contains the required software

  • ----knowledge-transfer set to previous if knowledge transfer from the previous chunk should be used
  • --suffix if the name of the imgstore contains anything after the datetime. Otherwise, just pass the empty string

idtrackerai repository version

62391f982531f3b8ab665d5f753bda169ed9dada

task spooler

https://github.com/justanhduc/task-spooler https://github.com/justanhduc/task-spooler/releases/tag/v1.3.0

rsync

# from the videos folder
REMOTE_FOLDER="/media/vibflysleep/Elements1/Users/Antonio/flyhostel_data/videos"
EXPERIMENT_FOLDER="2022-04-07_15-23-07"
rsync -arvz  -R --include "*/" --include "metadata.yaml" --include "*.conf" --include "*.npz" --include "*.json" --include "*.mp4" --include "*.png" --exclude "*" ./${EXPERIMENT_FOLDER}/ cv1:${REMOTE_FOLDER}

Analyse trajectories

The coordinates for every chunk are stored in the trajectories.npy file under

EXPERIMENT_FOLDER/idtrackerai/session_XXXXXX/trajectories/trajectories.npy

The entrypoint fh-copy will copy the trajectories.npy of each session to XXXXXX.npy on the EXPERIMENT_FOLDER, this way, besides the .json, .npz, .mp4 and .png files, a .npy becomes available for every chunk. The resulting copy should never be edited. Instead, one should edit them with python video annotator and copy the edited file in EXPERIMENT_FOLDER/idtrackerai/session_XXXXXX/trajectories/trajectories.npy back to the EXPERIMENT_FOLDER.

For example, from the videos folder:

fh-copy --imgstore-folder 2022-04-02_16-09-56 --analysis-folder 2022-04-02_16-09-56/idtrackerai/

will perform the EXPERIMENT_FOLDER/idtrackerai/session_XXXXXX/trajectories/trajectories.npy -> EXPERIMENT_FOLDER/XXXXXX.npy copies