adding inception_v3 extractor about ghiaseddin HOT 14 CLOSED

@yassersouri I have added the verification code, in commit cfcc256.
However, you should note that none of the codes in this commit are tested.

from ghiaseddin.

@yassersouri What do I exactly have to do for this issue? I should write the code for verifying the extractor? I don't have Caffe, so you should test it.

from ghiaseddin.

Inception v3 has input shape of bx3x299x299, what are we going to do about the difference in input image shape w.r.t. to other networks? Can the code handle this difference? (Are there any hard-coded dependencies on 244x244 image size?

from ghiaseddin.

You have to create the inception v3 and verify it with tensorflow. Do you have tensorflow installed?

We don't have any 224 hard coded anywhere.

Just add the network like googlenet.

from ghiaseddin.

No I don't have Tensorflow, neither. I'll write the code, you then run it.

from ghiaseddin.

You have made some changes to extractor (raw scale). I have to check whether the other extractors still work with this new thing.

from ghiaseddin.

thanks anyways.

from ghiaseddin.

@yassersouri Did you check this out? If you can't install Tensorflow, maybe I can find a way to install it in my machine (perhaps using Docker).

from ghiaseddin.

@yassersouri Also would you please add other issues in case I have to do anything else as soon as you can?

from ghiaseddin.

I will not invest time in this for now. I will come back to it when we will be training models using this extractor.

from ghiaseddin.

I have to fix this: currently logs of syntax errors

from ghiaseddin.

Lots have been fixed in c62e0bc

from ghiaseddin.

@yassersouri then what remains? Could you fine-tune Inception V3?

from ghiaseddin.

Nothing particularly is left. We have to verify that it actually is the same as the tensorflow version.

from ghiaseddin.