Model prediction is at 15 FPS, anyway to increase throughput?

[pangshengwei]

No description provided.

[JustinShenk]

Yes, we could batch images through the model. Are you using cpu only or also GPU acceleration?

[Neel7317]

how can we change FPS for Model prediction?

[JustinShenk]

Consider setting the save_fps parameter to a number 2-25. This will only compute on the nth frame.

[Neel7317]

thanks for answering but after changing save_fps parameter, i have to retrain the model ? and if i give save_fps=3 or 4 then will model get decrease its accuracy?

[JustinShenk]

`save_fps` just changes how many frames are skipped when running inference on the video. It doesn't change the model, only how it is applied to the video.

…

On Thu, Jun 17, 2021 at 11:59 AM Neelkanth ***@***.***> wrote: thanks for answering but after changing save_fps parameter, i have to retrain the model ? and if i give save_fps=3 or 4 then will model get decrease its accuracy? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#20 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACOLMZGQ6GMEXWXKLL4Z2BDTTHBONANCNFSM4355BPHQ> .

[Neel7317]

Thank you so much. I just found where to put that and it's working fine. and another doubt in my mind about whether we can have any solution to make this entire process fast? I have a GPU too but still it takes too much time for complete execution for just 1 minute video file. On Thu, Jun 17, 2021 at 4:29 PM Neelkanth Bhavnagarwala < ***@***.***> wrote:

…

means if i am using this below code then where do i make changes for save_fps can you tell me please ? def foremotion(): # fn contains the video file name (Ensure that you only upload one file) videofile = "D://RR/integration/in.mp4" # Face detection detector = FER(mtcnn=True) # Video predictions video = Video(videofile) # Output list of dictionaries raw_data = video.analyze(detector, display=False) # Convert to pandas for analysis df = video.to_pandas(raw_data) df = video.get_first_face(df) df = video.get_emotions(df) # Plot emotions fig = df.plot(figsize=(20, 10), fontsize=14).get_figure() # Filename for plot fig.savefig('static/my_figure.png') return render_template('try.html',img="static/my_figure.png") On Thu, Jun 17, 2021 at 3:57 PM Justin Shenk ***@***.***> wrote: > `save_fps` just changes how many frames are skipped when running inference > on the video. It doesn't change the model, only how it is applied to the > video. > > On Thu, Jun 17, 2021 at 11:59 AM Neelkanth ***@***.***> wrote: > > > thanks for answering but after changing save_fps parameter, i have to > > retrain the model ? and if i give save_fps=3 or 4 then will model get > > decrease its accuracy? > > > > — > > You are receiving this because you commented. > > Reply to this email directly, view it on GitHub > > <#20 (comment)>, > or > > unsubscribe > > < > https://github.com/notifications/unsubscribe-auth/ACOLMZGQ6GMEXWXKLL4Z2BDTTHBONANCNFSM4355BPHQ > > > > . > > > > — > You are receiving this because you commented. > Reply to this email directly, view it on GitHub > <#20 (comment)>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/APU3EWCJ6K7JBELVNRU4DYDTTHEX5ANCNFSM4355BPHQ> > . >

[JustinShenk]

It could be extended for this, maybe someone else will be kind enough to contribute this feature.

…

On Thu 17. Jun 2021 at 13:19 Neelkanth ***@***.***> wrote: Thank you so much. I just found where to put that and it's working fine. and another doubt in my mind about whether we can have any solution to make this entire process fast? I have a GPU too but still it takes too much time for complete execution for just 1 minute video file. On Thu, Jun 17, 2021 at 4:29 PM Neelkanth Bhavnagarwala < ***@***.***> wrote: > means if i am using this below code then where do i make changes for > save_fps can you tell me please ? > > def foremotion(): > # fn contains the video file name (Ensure that you only upload one file) > videofile = "D://RR/integration/in.mp4" > # Face detection > detector = FER(mtcnn=True) > # Video predictions > video = Video(videofile) > > # Output list of dictionaries > raw_data = video.analyze(detector, display=False) > > # Convert to pandas for analysis > df = video.to_pandas(raw_data) > df = video.get_first_face(df) > df = video.get_emotions(df) > > # Plot emotions > fig = df.plot(figsize=(20, 10), fontsize=14).get_figure() > # Filename for plot > fig.savefig('static/my_figure.png') > > return render_template('try.html',img="static/my_figure.png") > > > On Thu, Jun 17, 2021 at 3:57 PM Justin Shenk ***@***.***> > wrote: > >> `save_fps` just changes how many frames are skipped when running inference >> on the video. It doesn't change the model, only how it is applied to the >> video. >> >> On Thu, Jun 17, 2021 at 11:59 AM Neelkanth ***@***.***> wrote: >> >> > thanks for answering but after changing save_fps parameter, i have to >> > retrain the model ? and if i give save_fps=3 or 4 then will model get >> > decrease its accuracy? >> > >> > — >> > You are receiving this because you commented. >> > Reply to this email directly, view it on GitHub >> > <#20 (comment) >, >> or >> > unsubscribe >> > < >> https://github.com/notifications/unsubscribe-auth/ACOLMZGQ6GMEXWXKLL4Z2BDTTHBONANCNFSM4355BPHQ >> > >> > . >> > >> >> — >> You are receiving this because you commented. >> Reply to this email directly, view it on GitHub >> <#20 (comment)>, >> or unsubscribe >> < https://github.com/notifications/unsubscribe-auth/APU3EWCJ6K7JBELVNRU4DYDTTHEX5ANCNFSM4355BPHQ > >> . >> > — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#20 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACOLMZDAKOEATIGQSX4RDV3TTHK5DANCNFSM4355BPHQ> .