ACMtraingui

GUI to label frames for training of ACM-dlcdetect, by Arne Monsees

Installation

(You do not need to clone this repository.)

1. Install Anaconda
2. Start Anaconda Prompt (Windows) / terminal (linux) and navigate into repository directory
3. Create conda environment conda env create -f https://raw.githubusercontent.com/bbo-lab/ACM-traingui/master/environment.yml

Update

1. Start Anaconda Prompt (Windows) / terminal (linux) and navigate into repository directory
2. Update with conda env update -f https://raw.githubusercontent.com/bbo-lab/ACM-traingui/master/environment.yml --prune.

Running

1. Start Anaconda Prompt (Windows) / terminal (linux) and navigate into repository directory
2. Switch to environment conda activate bbo_acm-traingui
3. Run with python -m ACMtraingui [options ...]

Options

Assistant mode

Run with python -m ACMtraingui [base data directory]. This starts a GUI in drone mode, for the use by assistants with limited options to influence how the program runs and were it saves. This expects the following file structure:

[base data directory]/data/users/{user1,user2,...}/labeling_gui_cfg.py
[base data directory]/users/

{user1,user2,...} will be presented in a selection dialog on startup. Marking results will be placed in [base data directory]/users/

Master mode

Run with python -m ACMtraingui [configdir] --master. This starts a GUI in master mode. Only do this if you know what you are doing.

Check mode

Run with

python -m ACMtraingui [directory of labels.npz] --check [bbo_calibcam calibration npy] # to use specified npy or
python -m ACMtraingui [directory of labels.npz] --check [labeling_gui_cfg.py folder] # to use standardCalibrationFile from labeling_gui_cfg.py or
python -m ACMtraingui [directory of labels.npz] --check '-' # or
python -m ACMtraingui [directory of labels.npz] --check # to use backup labeling_gui_cfg.py in labels.npy folder. (Will often fail due to different paths between checker und labeler, as relative pathes are resolved, here).

This gives sorted text output of 3d and reprojections errors. Reporjection errors above 5-10px usually indicate errors in labeling and respective frames have to be checked.

Join mode

Run with python -m ACMtraingui [configdir of ACM-dlcdetect] --check [multiple directories containing labels.npz files] [--strict]. This joins all marked labels in the labels.npz files into the labels.npz file in the dlcdetect configuration. Marked labels overwrite existing labels framewise.

--strict only merges frames where all cameras have marked points.

TODO

• Document config
• Document sketch file (2d sketch of animal. If not presented, 3d wireframe is shown instead)
• Document model file