The close-up images are labeled on-the-fly using a customly programmed real-time version of DeepLabCut software. Once the body parts were labeled in an image, the frame number, the coordinates of the body parts and their likelihood were transmitted over the network to the host machine. The spatial relationship of individual real-time labeled body parts were used to define a given posture. Once the posture matches the target model, the EthoLoop system activates the RECO-Box. It is recommended to have DeepLabCut software running on a separate machine.
In the setup section we provide required steps using DeepLabCut software, however, you can use any other methods to detect body parts, as long as they are transmitted to the Host machine.
Have Ubuntu running on a separate machine and follow the guidelines described in the DeepLabCut github page for installation.
From your close up images choose ~300 images and label all the interested body part. Afterward train your network with DeepLabCut.
Download the source code for Realtime labeling of body parts. Afterwards install XIMEA SDK on your ubuntu machine and pass your config file for the training data set to the code.
Change the IP address in the source code to your Host machine IP address.
Using your favorite body parts design a simple filter that matches your target model in the loop of the python code. Send a boolean variable along with the frame number to the Host machine via the network.
On the host machine side where you have the EthoLoop software installed go to "Create Config File" and check on the "Reco-Box" option and choose Behaviour as your conditioning mode. Adjust RECO-Box parameters and have the real time body part labeling code running on the other machine.
Save the config file and run EthoLoop using the config file.
Start a free site - Visit site