Ensemble DGP Github Repo Link

Ensemble DGP Bash Script Link

Ensemble DGP Demo Link

How to use this analysis

The NeuroCAAS implementation of Ensembled DeepGraphPose trains a collection of N different networks on the same dataset, and predicts on the set of videos provided in the model folder. By default, running in training mode will also predict on the video found in the model folder's /videos directory. You can also run prediction only, as explained below.
TRAINING:
Args:
  -Input: (zip file) A zipped folder corresponding to a DLC/DGP model folder (with format {taskname}_{scorername}_{date}), containing the training frames to be used for training. This zipped file should include the folder itself. IMPORTANT: The folder name and zip archive name cannot contain spaces.
  -Config: (yaml) A YAML file containing the name of the directory that was zipped, and the DeepLabCut "myconfig" file. See template config for details.

mode	train
ensemble_size	9
nb_frames	5
jobnb	1
seed	5
videotype	mp4
testing	True
__duration__	220

PREDICTION:
Args:
  -Input: (video) A video or set of videos that you would like to run prediction on.
  -Config: (yaml) A YAML file containing the following model details:

mode	predict
modelpath	results/job__ensemble-dgp_2_20_90p_real/process_results/
modelnames	1 2 3 4 5
ensemble_size	11
nb_frames	52
seed	5
videotype	mp4
testing	True
__duration__	220

Note that "modelpath" is the location where the models you want to ensemble are located. If this is the output of a previous ensembling run, you can reference that ensemble by copying the path above, but giving the timestamp of your job instead: (results/{your timestamp}/process_results), and referencing the individual model folders within in the modelnames.

Outputs:
  -(zipfile) If training, full models fit from the data, as zip files results{modelnumber}.zip, and consensus traces, consensus_results_{videoname}.zip. If prediction only, returns only consensus traces.