moridinbg / face-it Goto Github PK

What steps will reproduce the problem?
1. Load example faces and non-faces
2. Start training
3. Observe the error rate changing

What is the expected output? What do you see instead?
The network should eventually reach a trained state. Instead it loops 
forever.
Using the same sets of training images may train it or may not. It seems 
this depends on the initial random weights

Please use labels and text to provide additional information.
A solution would be to automatically randomize the weights after a given 
number of iterations.

What steps will reproduce the problem?
1. Train the network with faces and non-faces
2. Load image for searching
3. Loads of false signals appear

What is the expected output? What do you see instead?
There should be at least one signal for a face and possibly several other 
false signals. Instead the false signals are 300-400

What steps will reproduce the problem?
1. Change the number of neurons in any layer to more than 400
2. Provide some training images
3. Start Training

What is the expected output? What do you see instead?
The network should start training with some info (error rate) being shown 
in the terminal. Instead it crashes with a backtrace

What steps will reproduce the problem?
1. Add some faces and non-faces in the network editor
2. Start training

What is the expected output? What do you see instead?
The training starts, some iterrations pass and their error is shown, but 
unexpectedly
the application segfaults, before reaching converged state

Please use labels and text to provide additional information.