Help on training the model, Please! Thanks.

[ghost]

Hi Alex,

I am using your Segnet to train the model following the procedures mentioned in "get started with segnet". I have got some errors after submit the following command. ./build/tools/caffe train -solver SegNet/Models/segnet_solver.prototxt

Error message:

Below is the steps I have done to build up the segnet caffe environment:

I built an image from the docker file through the following link. https://github.com/BVLC/caffe/blob/master/docker/gpu/Dockerfile

After I ran a container from the image, I placed CamVid and Models in the opt/caffe folder and modified the directory to absolute path in four files mentioned in the tutorial. For instance: "/SegNet/Models/segnet_train.prototxt" to "/opt/caffe/SegNet/Models/segnet_train.prototxt"

I am wondering if I have missed any procedures to make the model working. Please help me on this. Thank you very much in advance! Any advice is welcome.

[nathanin]

use segnet_solver.prototxt instead

[ghost]

@nathanin sorry, I do not mean segnet_train.prototxt. However, I mean segnet_solver,prototxt.

Please see the error after running it. Thanks for your time. Please let me know what do I need to do.

[nathanin]

Oh I just read the rest of your post.

You may need to build the Caffe version from this repository. dense_image_data_param is a protobuf definition related to one of the layers contributed from alexgkneall/callfe-segnet and is not present in the BVLC/caffe master branch.

You could replace these with normal data layers. But you still won't have the Upsample layers that are important for SegNet.

Please build the modified Caffe and try again.

Unless I'm mistaken?

[ghost]

I think you are right.

Do you mean I need to use "compilation with make" like mentioned in Caffe installation website to build the modified Caffe?

Are there any files from BVLC/caffe master branch I need to replace by alexgkneall/caffe-segnet files?

In addition, is there any path I need to modify in order to set up during the compilation?

When you do the segnet installation, do you clone the files from segnet github first and follow the caffe installation procedure ? Is that easier than building it from existing caffe environment?

Thanks for your time!

[ghost]

@nathanin

Do I need to clone everything from alexgkneall/caffe-segnet and overwrite them to /opt/caffe? then, I can compile caffe under /opt/caffe?