Open NikKaem opened 1 year ago
I guess the issue is that bpy.ops.wm.cgt_feature_detection_operator is bound to blenders window manager (mainly to show realtime updates) and while running blender headless this results in an error.
BlendArMocap/src/cgt_mediapipe/cgt_mp_detection_operator.py
if you check the operator, you'll see it uses the window manager. You can just use the operator as baseline to create your own operator, remove the modal, window manger stuff and just use a while loop in the execute
method.
If you have an issue with the opencv image display (idk if that's an issue in the docker as there is no window manager), you'll find a get stream
method in the operator. BlendArMocap/src/cgt_mediapipe/cgt_mp_core/cv_stream.py
-> You'll probably just overwrite the streams 'draw' method so it just passes on draw and should be good in this case.
So either overwrite in the add-ons operator directly or create an own operator :)
Quick outline:
class DOCKER_CGT_MP_modal_detection_operator(bpy.types.Operator):
# register the new operator with a custom id
...
def get_chain(self, stream) -> cgt_nodes.NodeChain:
...
return chain
def get_stream(self):
# overwrite the stream if necessary
return stream
def execute(self, context):
# here are some changes
self.user = context.scene.cgtinker_mediapipe # noqa
# init stream and chain
stream = self.get_stream()
self.node_chain = self.get_chain(stream)
if self.node_chain is None:
self.user.modal_active = False
return {'FINISHED'}
# memo skipped frames
self.memo = []
while self. overwritten_modal(context) == {'PASS_THROUGH'}:
pass
return {'OPERATOR_FINISHED'}
@staticmethod
def simple_smoothing(memo, cur):
...
return memo
def overwritten_modal(self, context):
""" Run detection as modal operation, finish with 'Q', 'ESC' or 'RIGHT MOUSE'. """
if self.user.detection_input_type == 'movie':
# get data
data, _frame = self.node_chain.nodes[0].update([], self.frame)
if data is None:
return self.cancel(context)
self.simple_smoothing(self.memo, data)
if self.frame % self.key_step == 0:
for node in self.node_chain.nodes[1:]:
node.update(self.memo, self.frame)
self.memo.clear()
self.frame += 1
return {'PASS_THROUGH'}
def cancel(self, context):
""" Upon finishing detection clear the handlers. """
del self.node_chain
return {'FINISHED'}
I just found the time to adapt the operator and it works like a charm. Haven't build a Docker container out of it yet but I can fully run the end-2-end process from the terminal now. Thank you so much! Is that something you would be interested in having as well?
You are welcome, glad it works!
Feel free to create a separate folder like src/docker
with a short readme so others which might need to run detection in a docker can use your implementation swell :)
@NikKaem It will be great if you would sare your solution. Thank you!
I'm currently trying to build a script that automates the whole end-2-end process with BlendARMocap
Minimal Code looks like this:
This works if I execute the terminal command
blender -P Script.py
. However, as soon as I try to keep blender in the background withblender -b -P Script.py
the detection part fails with the errorRuntimeError: expected class WM_OT_cgt_feature_detection_operator, function cancel to return None, not set
. I initially thought it was a problem with opening the detection window but even if I disable the drawing of the window the error persists. My ultimate goal is to run it all in a docker container which would be difficult if I need to show a frontend for it to work. Is there any workaround for that?