-
Notifications
You must be signed in to change notification settings - Fork 35
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
LMDB multidim labels #4
Comments
This is the incorrect way of specifying labels !! I urge you to please read the README instructions to rectify this !!
… On 10 Mar 2017, at 11:23, Alex Ter-Sarkisov ***@***.***> wrote:
I created train and test datasets with each entry looking like:
/Data/train/00001.png 347.0 178.0 408.0 283.0
where the four digits are bounding box values (x,y,h,w) of a bounding box. The label file was created, where every label entry looked like this:
<type 'numpy.int64'>
channels: 1
height: 1
width: 4
float_data: 347
float_data: 178
float_data: 408
float_data: 283
When training in caffe, I added a 4-neuron fully connected layer at the very end (before the error function) with relu activator. Is this the correct way of doing things? I get a huge error. Caffe reports a seemingly correct shape:
I0310 11:05:16.992735 4867 net.cpp:86] Creating Layer relu_fc8_cows
I0310 11:05:16.992741 4867 net.cpp:408] relu_fc8_cows <- fc8cows
I0310 11:05:16.992748 4867 net.cpp:369] relu_fc8_cows -> fc8cows (in-place)
I0310 11:05:16.993499 4867 net.cpp:124] Setting up relu_fc8_cows
I0310 11:05:16.993512 4867 net.cpp:131] Top shape: 1 4 (4)
—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub, or mute the thread.
|
I'm sorry, I don't understand why label values have to be between 0 and 255? If a label in my case is (x,y,h,w) bouding box parameters, for example? |
You just have to scale the values of your bounding box parameters to a range [0,255]. Internally since we are using a data layer for labels, Caffe rescales it by 255, so that will be your correct numbers then !! In a nutshell, you can have any range of your values, just scale it to [0,255] for inputting purposes !! |
But this means most numbers will be fractional. This is a problem. |
I created train and test datasets with each entry looking like:
/Data/train/00001.png 347.0 178.0 408.0 283.0
where the four digits are bounding box values (x,y,h,w) of a bounding box. The label file was created, where every label entry looked like this:
When training in caffe, I added a 4-neuron fully connected layer at the very end (before the error function) with relu activator. Is this the correct way of doing things? I get a huge error. Caffe reports a seemingly correct shape:
I0310 11:05:16.992735 4867 net.cpp:86] Creating Layer relu_fc8_cows
I0310 11:05:16.992741 4867 net.cpp:408] relu_fc8_cows <- fc8cows
I0310 11:05:16.992748 4867 net.cpp:369] relu_fc8_cows -> fc8cows (in-place)
I0310 11:05:16.993499 4867 net.cpp:124] Setting up relu_fc8_cows
I0310 11:05:16.993512 4867 net.cpp:131] Top shape: 1 4 (4)
The text was updated successfully, but these errors were encountered: