Hi all, I made a little video player using matplotlib. I need it to allow very good control over the playback speed (e.g. direction, frame-by-frame stepping and fast and slow). However, it's not very fast. Max frame rate I can achieve is 10-20 fps. I followed Basti's advice on speeding up plotting ( http://bastibe.de/2013-05-30-speeding-up-matplotlib.html). This resulted in a more than 2x improvement (from <5 to 10-20 fps), but I would like to reach 40-50 fps. The core code is something like this: self.im.set_data(self.video_frame) self.text.set_text('some text') self.ax.draw_artist(self.im) self.ax.draw_artist(self.text) self.im.figure.canvas.update() self.im.figure.canvas.flush_events() video_frame is a 200 x 250 array. I tried lowering dpi from 100 to 50, but the improvement is marginal. I saw that Harden ( http://www.swharden.com/wp/2013-04-15-fixing-slow-matplotlib-in-pythonxy/) recommended using TkAgg over Qt4Agg since TkAgg is supposed to be faster. But, TkAgg doesn't play nicely with figure.canvas.update(). I get the following error: AttributeError: 'FigureCanvasTkAgg' object has no attribute 'update' It works with Qt4Agg though. Does anyone have a fix for this? Or some general advice on how to speed up playback speed? I tried with both matplotlib 1.5.1 and the current version from github 2.0.0b4+2373.gb34c55d Best regards, Hjalmar
Instead of `canvas.update` call `self.im.figure.canvas.draw_idle()`. IIRC `update` is part of the API inherited from Qt, not part of the API we ensure that all of the canvas objects have. Tom On Fri, Oct 7, 2016 at 4:48 PM Hjalmar Turesson <htu...@gm...> wrote: > Hi all, > > > I made a little video player using matplotlib. I need it to allow very > good control over the playback speed (e.g. direction, frame-by-frame > stepping and fast and slow). > > However, it's not very fast. Max frame rate I can achieve is 10-20 fps. > I followed Basti's advice on speeding up plotting ( > http://bastibe.de/2013-05-30-speeding-up-matplotlib.html). > This resulted in a more than 2x improvement (from <5 to 10-20 fps), but I > would like to reach 40-50 fps. > > The core code is something like this: > > self.im.set_data(self.video_frame) > > self.text.set_text('some text') > > self.ax.draw_artist(self.im) > self.ax.draw_artist(self.text) > self.im.figure.canvas.update() > self.im.figure.canvas.flush_events() > > video_frame is a 200 x 250 array. I tried lowering dpi from 100 to 50, > but the improvement is marginal. > > I saw that Harden ( > http://www.swharden.com/wp/2013-04-15-fixing-slow-matplotlib-in-pythonxy/) > recommended using TkAgg over Qt4Agg since TkAgg is supposed to be faster. > But, TkAgg doesn't play nicely with figure.canvas.update(). I get the > following error: > AttributeError: 'FigureCanvasTkAgg' object has no attribute 'update' > It works with Qt4Agg though. > > Does anyone have a fix for this? Or some general advice on how to speed up > playback speed? > > I tried with both matplotlib 1.5.1 and the current version from > github 2.0.0b4+2373.gb34c55d > > Best regards, > Hjalmar > > ------------------------------------------------------------------------------ > Check out the vibrant tech community on one of the world's most > engaging tech sites, SlashDot.org! http://sdm.link/slashdot > _______________________________________________ > Matplotlib-users mailing list > Mat...@li... > https://lists.sourceforge.net/lists/listinfo/matplotlib-users >
That works fine. And it explains why update() only worked with the Qt backends (I tried all). The speed is still not super impressive though (~20 fps), but I think I will just start skipping frames when playing at above 20 fps. Thanks, Hjalmar On Fri, Oct 14, 2016 at 10:37 AM, Thomas Caswell <tca...@gm...> wrote: > Instead of `canvas.update` call `self.im.figure.canvas.draw_idle()`. > > IIRC `update` is part of the API inherited from Qt, not part of the API we > ensure that all of the canvas objects have. > > Tom > > On Fri, Oct 7, 2016 at 4:48 PM Hjalmar Turesson <htu...@gm...> > wrote: > >> Hi all, >> >> >> I made a little video player using matplotlib. I need it to allow very >> good control over the playback speed (e.g. direction, frame-by-frame >> stepping and fast and slow). >> >> However, it's not very fast. Max frame rate I can achieve is 10-20 fps. >> I followed Basti's advice on speeding up plotting ( >> http://bastibe.de/2013-05-30-speeding-up-matplotlib.html). >> This resulted in a more than 2x improvement (from <5 to 10-20 fps), but I >> would like to reach 40-50 fps. >> >> The core code is something like this: >> >> self.im.set_data(self.video_frame) >> >> self.text.set_text('some text') >> >> self.ax.draw_artist(self.im) >> self.ax.draw_artist(self.text) >> self.im.figure.canvas.update() >> self.im.figure.canvas.flush_events() >> >> video_frame is a 200 x 250 array. I tried lowering dpi from 100 to 50, >> but the improvement is marginal. >> >> I saw that Harden (http://www.swharden.com/wp/2013-04-15-fixing-slow- >> matplotlib-in-pythonxy/) recommended using TkAgg over Qt4Agg since TkAgg >> is supposed to be faster. >> But, TkAgg doesn't play nicely with figure.canvas.update(). I get the >> following error: >> AttributeError: 'FigureCanvasTkAgg' object has no attribute 'update' >> It works with Qt4Agg though. >> >> Does anyone have a fix for this? Or some general advice on how to speed >> up playback speed? >> >> I tried with both matplotlib 1.5.1 and the current version from >> github 2.0.0b4+2373.gb34c55d >> >> Best regards, >> Hjalmar >> ------------------------------------------------------------ >> ------------------ >> Check out the vibrant tech community on one of the world's most >> engaging tech sites, SlashDot.org! http://sdm.link/slashdot______ >> _________________________________________ >> Matplotlib-users mailing list >> Mat...@li... >> https://lists.sourceforge.net/lists/listinfo/matplotlib-users >> >
To push much past 20Hz you will want to look into blitting. See http://matplotlib.org/devdocs/api/animation_api.html for a rough introduction on how to use blitting (and see the animation code for an example of handling all of the corner cases). Tom On Fri, Oct 14, 2016 at 5:29 PM Hjalmar Turesson <htu...@gm...> wrote: > That works fine. And it explains why update() only worked with the Qt > backends (I tried all). The speed is still not super impressive though (~20 > fps), but I think I will just start skipping frames when playing at above > 20 fps. > > Thanks, > Hjalmar > > On Fri, Oct 14, 2016 at 10:37 AM, Thomas Caswell <tca...@gm...> > wrote: > > Instead of `canvas.update` call `self.im.figure.canvas.draw_idle()`. > > IIRC `update` is part of the API inherited from Qt, not part of the API we > ensure that all of the canvas objects have. > > Tom > > On Fri, Oct 7, 2016 at 4:48 PM Hjalmar Turesson <htu...@gm...> > wrote: > > Hi all, > > > I made a little video player using matplotlib. I need it to allow very > good control over the playback speed (e.g. direction, frame-by-frame > stepping and fast and slow). > > However, it's not very fast. Max frame rate I can achieve is 10-20 fps. > I followed Basti's advice on speeding up plotting ( > http://bastibe.de/2013-05-30-speeding-up-matplotlib.html). > This resulted in a more than 2x improvement (from <5 to 10-20 fps), but I > would like to reach 40-50 fps. > > The core code is something like this: > > self.im.set_data(self.video_frame) > > self.text.set_text('some text') > > self.ax.draw_artist(self.im) > self.ax.draw_artist(self.text) > self.im.figure.canvas.update() > self.im.figure.canvas.flush_events() > > video_frame is a 200 x 250 array. I tried lowering dpi from 100 to 50, > but the improvement is marginal. > > I saw that Harden ( > http://www.swharden.com/wp/2013-04-15-fixing-slow-matplotlib-in-pythonxy/) > recommended using TkAgg over Qt4Agg since TkAgg is supposed to be faster. > But, TkAgg doesn't play nicely with figure.canvas.update(). I get the > following error: > AttributeError: 'FigureCanvasTkAgg' object has no attribute 'update' > It works with Qt4Agg though. > > Does anyone have a fix for this? Or some general advice on how to speed up > playback speed? > > I tried with both matplotlib 1.5.1 and the current version from > github 2.0.0b4+2373.gb34c55d > > Best regards, > Hjalmar > > ------------------------------------------------------------------------------ > Check out the vibrant tech community on one of the world's most > engaging tech sites, SlashDot.org! http://sdm.link/slashdot > _______________________________________________ > Matplotlib-users mailing list > Mat...@li... > https://lists.sourceforge.net/lists/listinfo/matplotlib-users > > >
Thanks for the help (and sorry for the super-late reply). I just tried blitting, but it doesn't really help. I cannot exceed 20 fps. The problem is that I'm plotting data from a video file, so that for each frame I need to draw a new array (not just a foreground). I think its just to much data. Thanks, Hjalmar On Sat, Jan 7, 2017 at 3:36 PM, Thomas Caswell <tca...@gm...> wrote: > To push much past 20Hz you will want to look into blitting. See > http://matplotlib.org/devdocs/api/animation_api.html for a rough > introduction on how to use blitting (and see the animation code for an > example of handling all of the corner cases). > > Tom > > On Fri, Oct 14, 2016 at 5:29 PM Hjalmar Turesson <htu...@gm...> > wrote: > >> That works fine. And it explains why update() only worked with the Qt >> backends (I tried all). The speed is still not super impressive though (~20 >> fps), but I think I will just start skipping frames when playing at above >> 20 fps. >> >> Thanks, >> Hjalmar >> >> On Fri, Oct 14, 2016 at 10:37 AM, Thomas Caswell <tca...@gm...> >> wrote: >> >> Instead of `canvas.update` call `self.im.figure.canvas.draw_idle()`. >> >> IIRC `update` is part of the API inherited from Qt, not part of the API >> we ensure that all of the canvas objects have. >> >> Tom >> >> On Fri, Oct 7, 2016 at 4:48 PM Hjalmar Turesson <htu...@gm...> >> wrote: >> >> Hi all, >> >> >> I made a little video player using matplotlib. I need it to allow very >> good control over the playback speed (e.g. direction, frame-by-frame >> stepping and fast and slow). >> >> However, it's not very fast. Max frame rate I can achieve is 10-20 fps. >> I followed Basti's advice on speeding up plotting ( >> http://bastibe.de/2013-05-30-speeding-up-matplotlib.html). >> This resulted in a more than 2x improvement (from <5 to 10-20 fps), but I >> would like to reach 40-50 fps. >> >> The core code is something like this: >> >> self.im.set_data(self.video_frame) >> >> self.text.set_text('some text') >> >> self.ax.draw_artist(self.im) >> self.ax.draw_artist(self.text) >> self.im.figure.canvas.update() >> self.im.figure.canvas.flush_events() >> >> video_frame is a 200 x 250 array. I tried lowering dpi from 100 to 50, >> but the improvement is marginal. >> >> I saw that Harden (http://www.swharden.com/wp/2013-04-15-fixing-slow- >> matplotlib-in-pythonxy/) recommended using TkAgg over Qt4Agg since TkAgg >> is supposed to be faster. >> But, TkAgg doesn't play nicely with figure.canvas.update(). I get the >> following error: >> AttributeError: 'FigureCanvasTkAgg' object has no attribute 'update' >> It works with Qt4Agg though. >> >> Does anyone have a fix for this? Or some general advice on how to speed >> up playback speed? >> >> I tried with both matplotlib 1.5.1 and the current version from >> github 2.0.0b4+2373.gb34c55d >> >> Best regards, >> Hjalmar >> ------------------------------------------------------------ >> ------------------ >> Check out the vibrant tech community on one of the world's most >> engaging tech sites, SlashDot.org! http://sdm.link/slashdot______ >> _________________________________________ >> Matplotlib-users mailing list >> Mat...@li... >> https://lists.sourceforge.net/lists/listinfo/matplotlib-users >> >> >>
I am wondering if the "optimizations" you have are actually slowing you down. I have never found myself needing to flush_events() or call update() like that. Or to draw the artists like you are doing. Without seeing more of the code, it is hard to judge. Have you tried using "runsnakerun" to profile your code to find out what is actually slowing it down? I can get 60-70Hz without even trying for 3d plotting animations, and that is probably more computation than this. On Tue, Apr 18, 2017 at 4:50 PM, Hjalmar Turesson <htu...@gm...> wrote: > Thanks for the help (and sorry for the super-late reply). > > I just tried blitting, but it doesn't really help. I cannot exceed 20 fps. > The problem is that I'm plotting data from a video file, so that for each > frame I need to draw a new array (not just a foreground). I think its just > to much data. > > Thanks, > Hjalmar > > On Sat, Jan 7, 2017 at 3:36 PM, Thomas Caswell <tca...@gm...> wrote: > >> To push much past 20Hz you will want to look into blitting. See >> http://matplotlib.org/devdocs/api/animation_api.html for a rough >> introduction on how to use blitting (and see the animation code for an >> example of handling all of the corner cases). >> >> Tom >> >> On Fri, Oct 14, 2016 at 5:29 PM Hjalmar Turesson <htu...@gm...> >> wrote: >> >>> That works fine. And it explains why update() only worked with the Qt >>> backends (I tried all). The speed is still not super impressive though (~20 >>> fps), but I think I will just start skipping frames when playing at above >>> 20 fps. >>> >>> Thanks, >>> Hjalmar >>> >>> On Fri, Oct 14, 2016 at 10:37 AM, Thomas Caswell <tca...@gm...> >>> wrote: >>> >>> Instead of `canvas.update` call `self.im.figure.canvas.draw_idle()`. >>> >>> IIRC `update` is part of the API inherited from Qt, not part of the API >>> we ensure that all of the canvas objects have. >>> >>> Tom >>> >>> On Fri, Oct 7, 2016 at 4:48 PM Hjalmar Turesson <htu...@gm...> >>> wrote: >>> >>> Hi all, >>> >>> >>> I made a little video player using matplotlib. I need it to allow very >>> good control over the playback speed (e.g. direction, frame-by-frame >>> stepping and fast and slow). >>> >>> However, it's not very fast. Max frame rate I can achieve is 10-20 fps. >>> I followed Basti's advice on speeding up plotting ( >>> http://bastibe.de/2013-05-30-speeding-up-matplotlib.html). >>> This resulted in a more than 2x improvement (from <5 to 10-20 fps), but >>> I would like to reach 40-50 fps. >>> >>> The core code is something like this: >>> >>> self.im.set_data(self.video_frame) >>> >>> self.text.set_text('some text') >>> >>> self.ax.draw_artist(self.im) >>> self.ax.draw_artist(self.text) >>> self.im.figure.canvas.update() >>> self.im.figure.canvas.flush_events() >>> >>> video_frame is a 200 x 250 array. I tried lowering dpi from 100 to 50, >>> but the improvement is marginal. >>> >>> I saw that Harden (http://www.swharden.com/wp/20 >>> 13-04-15-fixing-slow-matplotlib-in-pythonxy/) recommended using TkAgg >>> over Qt4Agg since TkAgg is supposed to be faster. >>> But, TkAgg doesn't play nicely with figure.canvas.update(). I get the >>> following error: >>> AttributeError: 'FigureCanvasTkAgg' object has no attribute 'update' >>> It works with Qt4Agg though. >>> >>> Does anyone have a fix for this? Or some general advice on how to speed >>> up playback speed? >>> >>> I tried with both matplotlib 1.5.1 and the current version from >>> github 2.0.0b4+2373.gb34c55d >>> >>> Best regards, >>> Hjalmar >>> ------------------------------------------------------------ >>> ------------------ >>> Check out the vibrant tech community on one of the world's most >>> engaging tech sites, SlashDot.org! http://sdm.link/slashdot______ >>> _________________________________________ >>> Matplotlib-users mailing list >>> Mat...@li... >>> https://lists.sourceforge.net/lists/listinfo/matplotlib-users >>> >>> >>> > > ------------------------------------------------------------ > ------------------ > Check out the vibrant tech community on one of the world's most > engaging tech sites, Slashdot.org! http://sdm.link/slashdot > _______________________________________________ > Matplotlib-users mailing list > Mat...@li... > https://lists.sourceforge.net/lists/listinfo/matplotlib-users > >