search

FORTH-ModelBasedTracker / PyOpenPose

Python bindings for the Openpose library
BSD 3-Clause "New" or "Revised" License
287 stars 80 forks source link

Which parameter controls GPU memory consumption? #42

Closed AlexLuya closed 5 years ago

AlexLuya commented 6 years ago

Hello,I am using PyOpenpose with below settings:

       self.poseSize = (656, 368)
        self.faceHandSize = (240, 240)
        self.outSize = (640, 480)
        self.modelType = "COCO"
        self.logLevel = 0
        self.downloadHeatmaps = False
        self.detectFace = False
        self.detectHands = False

Same python scripts+same settings as above, under my laptop with a nvidia 1070 card,python wrapped openpose process consumed 1516MB GPU memory, but under a server with nvidia 1080Ti card,openpose process consumed 7883MB GPU memory, where does this big memory consumption comef from?Any parameter to specify GPU memory consumption? (I have asked in openpose issue list:https://github.com/CMU-Perceptual-Computing-Lab/openpose/issues/526,but openpose guys said it isn't their problem)

padeler commented 6 years ago

This is not a PyOpenPose related issue. That said, try changing (reducing) the input resolution when initializing the PyOpenPose class and see how the memory consumption changes.