- Download model.ckpt from https://gvv-assets.mpi-inf.mpg.de/EventHands/ to
live_demo/
. - In DV, clear the graph, add the event camera node, add a network node. Connect events in the camera output to the network node input.
Note:
You can save the DV graph as .xml
and load it later instead of recreating it every time.
Note:
You can use a prerecorded .aedat4
file instead of an event camera. Instead of an event camera node, use the aedat4 reader node.
- Compile the native module for the live prediction code. More info is in the corresponding README.md.
- Run
live_predict.py
. This will capture DV event stream via network and output the predictions to/tmp/preds
FIFO. - Run the event simulator compiled from
evsim_live
using the predicted outputs at/tmp/preds
(run atevsim_joints
working directory where all the simulator assets are):
../evsim_live_build/evsim -p fifo:///tmp/preds
You should be able to see the live predictions on the screen.