search

DmitryUlyanov / deep-image-prior

Image restoration with neural networks but without learning.
https://dmitryulyanov.github.io/deep_image_prior
Other
7.79k stars 1.42k forks source link

Question: is it possible to use it on Android device? #37

Open AndroidDeveloperLB opened 5 years ago

AndroidDeveloperLB commented 5 years ago

does it take a lot of computing power to use this ? Does it require training ? of a lot of images? Can it have the training data built into the app? Without any server?

pablodz commented 5 years ago

does it take a lot of computing power to use this ? Does it require training ? of a lot of images? Can it have the training data built into the app? Without any server?

Yes, It's a lot of GPU

AndroidDeveloperLB commented 5 years ago

So, supposed I have 1000 images of 4K quality, how much time could it take to train on PC vs one of the newest smartphones out there over the past year (say, Pixel 2) ?

IEWbgfnYDwHRoRRSKtkdyMDUzgdwuBYgDKtDJWd commented 5 years ago

dude, it will take literally a million years. stop asking stupid questions and learn about the basics.

AndroidDeveloperLB commented 5 years ago

@IEWbgfnYDwHRoRRSKtkdyMDUzgdwuBYgDKtDJWd It's not stupid. Smartphone can do long work. Not at the same level as PC, but still...

pablodz commented 5 years ago

So, supposed I have 1000 images of 4K quality, how much time could it take to train on PC vs one of the newest smartphones out there over the past year (say, Pixel 2) ?

GPU of PC is the reason why PC can works normaly with this kind of processing. However, The smartphones are designed to just look nice.

AndroidDeveloperLB commented 5 years ago

@ZurMaD Do you think that one day, smarphones will reach the same performance as PC (on the same time era) ? I think they are becoming more and more powerful, faster than PCs, so maybe it is possible. I think it was said a year ago (or so) by ARM, that smartphones have reached a level comparable to consoles, in some way. Is it true?

pablodz commented 5 years ago

@ZurMaD Do you think that one day, smarphones will reach the same performance as PC (on the same time era) ? I think they are becoming more and more powerful, faster than PCs, so maybe it is possible. I think it was said a year ago (or so) by ARM, that smartphones have reached a level comparable to consoles, in some way. Is it true?

In reality, big companies are building processes in the cloud, making smartphones more dependent with Internet, because they can optimize. As Facebook Face Detector in the app.

AndroidDeveloperLB commented 5 years ago

@ZurMaD I see, but what is Tensor Lite, then? It's not meant for machine learning, right on the smartphone? Suppose you do all the training on very good PCs, can a smartphone have what was learned right on the app, without the need of Internet connection?

pablodz commented 5 years ago

@ZurMaD I see, but what is Tensor Lite, then? It's not meant for machine learning, right on the smartphone? Suppose you do all the training on very good PCs, can a smartphone have what was learned right on the app, without the need of Internet connection?

Yes, it's totally possible but not recommended because it makes the dispositives too expensive.

AndroidDeveloperLB commented 5 years ago

@ZurMaD What do you mean by "dispositives too expensive" ? You mean it will take a lot of space in the storage?

pablodz commented 5 years ago

@ZurMaD What do you mean by "dispositives too expensive" ? You mean it will take a lot of space in the storage?

No, I'm saying that GPUs made devices more expensive.

AndroidDeveloperLB commented 5 years ago

Are you talking about PC? Because smartphones are just what they are... They keep getting replaced by new ones all the time, one small step after another.

magneter commented 5 years ago

Actually ,we should evaluate the model to decide wether it is suitable running on the phone and the problem we try to solve ,the accuracy ,precision and fps....can not be ignored. You can optimize ,remove some branches, reduce the kernels , quantize weights .... Lots of models have been running on the phone now to sovle problems. But ,if the problem needs very high precision, and high fps, especially in efficiency ,you had better run it on the GPUs. For example, this model is not suitable to be deployed on phone.