SegNet and NVIDIA DIGITS

[Almo1982]

Hi all,

did anybody succesfully trained SegNet with NVIDIA DIGITS?

As far as i understand only the "official" BVLC/caffe is supported from DIGITS but isn't there any way to use SegNet within DIGITS?

Actually the error i get is that when i copy the train net-definition into the custom Network area and try to start training it says "no Name bn_param in caffe.Layerparam". I recognized that there is no bn_param.cpp in the Layer Folder of the official caffe build, what explains this error. In other Forum i read that caffe supports "Batch_norm_param"-layer, but if i replace bn_param with Batch_norm_param another error occurs "no Name shift_filler found".

Is there any Chance to copy the missing (bn_param) layer into the Folders of the official caffe build? i guess ist not that easy, i tried a lot of things but nothing worked so far.

Or is there a possibility to tell DIGITS not to use the "Standard" caffe-build but the "caffe-segnet"-build? I added the caffe-segnet build path to my bashrc, didnt work too...

Any help would be appreciated!

thank you.

Regards Stefan

[eliethesaiyan]

@alexgkendall ,is there anyway we can use digitis with segnet?

[eliethesaiyan]

@Almo1982 i managed to run digits on caffe-segnet on digits but i have been able to fully implement the inference visualization yet...you basically have to ignore the checking of those layers in digits/config/caffe.py since they are present but digits doesn't recognize them

[mcamila777]

@Almo1982 did you find any solution? I have the same issue when tryng to use SegNet with DIGITS:

47:3 : Message type "caffe.LayerParameter" has no field named "bn_param".

[eliethesaiyan]

@mcamila777 can you give the whole traceback?

[mcamila777]

@eliethesaiyan I'm using the caffe-segnet version, I can train the segnet network from the cmd normally, but when I try with DIGITS i get the error: Message type "caffe.LayerParameter" has no field named "bn_param". I checked the caffe-segnet layers files and in my caffe-segnet this "bn_param" is set , while in the caffe original repo it is not. I dont untdertand why this parameter is recognized for the caffe normal training in cmd but not in DIGITS....

I want to use DIGITS for visualization of the test segmented data, as I donwloaded different segnet pre-trained examples and I want to see the performance with images of the web-demo which weight are available, because it was trained with a huge amount of images. (With the tutorial example I could visualizate the test segmented images for a model that I trained starting with the VGG16 inicialization, but as it is trained with arround 300 images the results are not pretty good.)

[eliethesaiyan]

@mcamila777 ,can you give the full stack trace?i think you are right,NVIDIA support the Berkley version of caffe,i think bn_param and DenseImageLayer were added by the author of caffe-segnet,for me i loaded my dataset into lmdb database(meaning that there is no batch_norm),you can do this easily in nvdia,and used the Data Layer instead of DenseImageDataLayer,after that you can follow this example ,i didn't manage to run it either