عبير ________
2011-04-26 00:07:26 UTC
Hi everybody,
I have a project depends on road sign detection(not recognition) from image
of scene.
Firstly, i just need to detect if the input image is circle, triangle road
sign or not!
To accomplish this goal, i created neural net with 3 layers.
Input layer=900 neurons (because the image is 30Ã30 pixels)
Output layer=3 (1 0 0 for circle, 0 1 0 for triangle)
hidden layer = 1500
const float connection_rate = 1;
const float learning_rate = 0.7;
I used standard create function.
Also, I used ( float train_epoch(const training_data & data)
the data has been read from file.data then it has stored in object of
training_data).
The image is 30Ã30 pixels, it has been converted to zeros & ones file
The format of training file is:
-------------------------------------
60 900 3
(Here, zero&one image in one line not like matrix)
1 0 0 (if the input was circle)
----------------------------------------
As we see input is 60 image
30 circles
30 triangle
The problem is:
when i enter image from the training file as an input to run net
i get correct output, but when i do test on a new circle for example, I get
often wrong output!!!!
I need to know where is the error to fix it!
is it in format of training file?
is it in number of output neurons?
is it in number of training inputs?
PLZ I need to help, I have not made ââany progress with him weeks ago
NOTE:
If it is necessary I can put the code that I wrote here.
I have a project depends on road sign detection(not recognition) from image
of scene.
Firstly, i just need to detect if the input image is circle, triangle road
sign or not!
To accomplish this goal, i created neural net with 3 layers.
Input layer=900 neurons (because the image is 30Ã30 pixels)
Output layer=3 (1 0 0 for circle, 0 1 0 for triangle)
hidden layer = 1500
const float connection_rate = 1;
const float learning_rate = 0.7;
I used standard create function.
Also, I used ( float train_epoch(const training_data & data)
the data has been read from file.data then it has stored in object of
training_data).
The image is 30Ã30 pixels, it has been converted to zeros & ones file
The format of training file is:
-------------------------------------
60 900 3
(Here, zero&one image in one line not like matrix)
1 0 0 (if the input was circle)
----------------------------------------
As we see input is 60 image
30 circles
30 triangle
The problem is:
when i enter image from the training file as an input to run net
i get correct output, but when i do test on a new circle for example, I get
often wrong output!!!!
I need to know where is the error to fix it!
is it in format of training file?
is it in number of output neurons?
is it in number of training inputs?
PLZ I need to help, I have not made ââany progress with him weeks ago
NOTE:
If it is necessary I can put the code that I wrote here.