search

MarekKowalski / DeepAlignmentNetwork

A deep neural network for face alignment
MIT License
510 stars 136 forks source link

question about the ImageDemo #39

Open JWSunny opened 6 years ago

JWSunny commented 6 years ago

hello： when i test the ImageDemo.py, i find that a picture of 600*640 pixes need 17s on cpu, and 10s on gpu; However, this is not the case in the paper. Is the processing time normal? Time is mainly spent on a function of theano, can time continue to be optimized? thanks，best wishes

MarekKowalski commented 6 years ago

Hi,

That is definitely not normal. Have you tried running other theano DNNs on your machine? What hardware setup do you have? Also, the processing time should be independent of the image size (all images are cropped and resized to the same 112x112 size).

Thanks,

Marek

JWSunny commented 6 years ago

Hi Marek： I mainly run un ImageDemo.py, I , I tested a picture, the timing is started by by model.loadNetwork, rk, the time is mainly spent on the he model.processcessImg function, the picture will be processed to 112112, but in in FaceAlignment.processcessImg the he self.generate_net_network_output function took a long time. The inputImg is 112112. Is my way of calculating time wrong? Thanks，best wishes Sunny

MarekKowalski commented 6 years ago

Hi,

The runtime of the first image that is processed may be a bit longer. Please calculate the time starting from the second image you process. Also, do not include the model.loadNetwork as this is done only once, not per-image.

Best regards,

Marek

youngallien commented 5 years ago

hi, i test my picture on RTX2060,and the time is 1.5s per image,it is too slow to run real time.how can i opimiter the code? thanks

MarekKowalski commented 5 years ago

That's definitely slower then it should be. Are you sure it's running on the GPU? Theano should tell you in the first line it prints whether it's using the GPU or not.