Re: Feature extraction: Files(000003.log, CURRENT, LOCK, LOG, MANIFEST-000002)

[Jan]

These files make up a database (LMDB or LevelDB, not sure which one, you should know). You should never access them directly, but only through the LMDB/LevelDB APIs. The "database" is usually identified with the name of the directory these files are in.

Jan


Am Donnerstag, 24. März 2016 13:39:50 UTC+1 schrieb monjoy saha:

After feature extraction I got following file

000003.log, CURRENT, LOCK, LOG, MANIFEST-000002

Can anyone kindly explain what does these files mean? which file is required for further processing in MATLAB?

[Jan]

I don't know that, I have never used the matlab interface. I can only help with pycaffe.

Jan


Am Mittwoch, 30. März 2016 10:08:02 UTC+2 schrieb monjoy saha:

@Jan, I have one request please help me.

I am using imagenet model in my own dataset. I want to see the result after each stage like if I want to see the result after 1st convolution then what will be MATLAB code? please help.

[Ahmed Ibrahim]

every layer has a top blob which you can extract easily using matlab script like that

 conv1output= net.blobs('conv1').get_data();

where net is the network you provided the input to.

[Ahmed Ibrahim]

you are welcome , we are here to help each other , hope you help someone in the future
you did not provide input to the network you should do something like

res = net.forward({data});
prob = res{1};

On Tue, Apr 5, 2016 at 4:26 AM, monjoy saha <monj...@gmail.com> wrote:

@ Ahmed Ibrahim & Jan Thanks a lot. Today I completed training of images in my own net. Received lots of suggestions from this group hope I ''ll get same help in future.

I successfully trained my network and the I achieved accuracy = 0.842145 and loss = 0.494893 (* 1 = 0.494893 loss). Now I want to extract the data from 1st convolution blob. so I did this in MATLAB command window:

model = './examples/mi/deploy.prototxt';
 weights = './examples/mi/recent_train_iter_30000.caffemodel';
caffe.set_mode_gpu();
net = caffe.Net(model, weights, 'test');
conv1output= net.blobs('conv1').get_data();

In the MATLAB workplace conv1output is creating with all ZEROS. in the same way other layer also showing all ZEROS please suggest what to do in this case.

On Thursday, 24 March 2016 13:39:50 UTC+1, monjoy saha wrote:

After feature extraction I got following file

000003.log, CURRENT, LOCK, LOG, MANIFEST-000002

Can anyone kindly explain what does these files mean? which file is required for further processing in MATLAB?

--
You received this message because you are subscribed to a topic in the Google Groups "Caffe Users" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/caffe-users/WVMosMK0Twg/unsubscribe.
To unsubscribe from this group and all its topics, send an email to caffe-users...@googlegroups.com.
To post to this group, send email to caffe...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/caffe-users/f64fe98c-30ed-4be4-9ce6-6a378d67341c%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

Ahmed Ibrahim
PhD Student , Virginia Tech.
Electrical and Computer Engineering

[Ahmed Ibrahim]

Data should be the input you would like to see the conv for . an input image but in format accepted by caffe ( the Matlab ddmo code that comes with caffe contains a function called prepare_image that you can look at)

On Apr 5, 2016 6:36 AM, "monjoy saha" <monj...@gmail.com> wrote:

@Ahmed Ibrahim

Thanks a lot. In the below code how I can I define "data"? hope the code will be like below

model = './examples/mi/deploy.prototxt';
 weights = './examples/mi/recent_train_iter_30000.caffemodel';
caffe.set_mode_gpu();
net = caffe.Net(model, weights, 'test');

res = net.forward({data});
prob = res{1};

conv1output= net.blobs('conv1').get_data();

am I right? Please suggest.

On Thursday, 24 March 2016 13:39:50 UTC+1, monjoy saha wrote:

After feature extraction I got following file

000003.log, CURRENT, LOCK, LOG, MANIFEST-000002

Can anyone kindly explain what does these files mean? which file is required for further processing in MATLAB?

--
You received this message because you are subscribed to a topic in the Google Groups "Caffe Users" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/caffe-users/WVMosMK0Twg/unsubscribe.
To unsubscribe from this group and all its topics, send an email to caffe-users...@googlegroups.com.
To post to this group, send email to caffe...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/caffe-users/b605979a-6aff-46c2-8b40-939e00a35e40%40googlegroups.com.

[Ahmed Ibrahim]

This should be the same mean file you used during training

On Apr 5, 2016 7:23 AM, "monjoy saha" <monj...@gmail.com> wrote:

Thank you.

In classification_demo, under prepare_image(im) I can see "ilsvrc_2012_mean.mat" has been used. In the drescription it is mentioned that "ilsvrc_2012_mean.mat" contains mean_data that is already in W x H x C with BGR channels.

My question is that how can I prepare this .mat file? Is it same file of mean_imagenet.binaryproto and only converted into .mat file? or its something else?

Please suggest.

On Thursday, 24 March 2016 13:39:50 UTC+1, monjoy saha wrote:

After feature extraction I got following file

000003.log, CURRENT, LOCK, LOG, MANIFEST-000002

Can anyone kindly explain what does these files mean? which file is required for further processing in MATLAB?

--
You received this message because you are subscribed to a topic in the Google Groups "Caffe Users" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/caffe-users/WVMosMK0Twg/unsubscribe.
To unsubscribe from this group and all its topics, send an email to caffe-users...@googlegroups.com.
To post to this group, send email to caffe...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/caffe-users/b8f20e74-640c-4619-a3d7-5cf8afb876cc%40googlegroups.com.

[r...@minerva.kgi.edu]

How can I use the features on a python code without using the levelDB API? Is there a way to copy the features into something like a numpy array?

[Hamed Karimi]

Dear Saha,
Thanks for your question. I wanted to extract features. I got the files as you obtained. What are these files? How can I use them for further process?
Thanks in advance.