search

mbeyeler / opencv-python-blueprints

M. Beyeler (2015). OpenCV with Python Blueprints: Design and develop advanced computer vision projects using OpenCV with Python, Packt Publishing Ltd., ISBN 978-178528269-0.
GNU General Public License v3.0
291 stars 184 forks source link

Could not find data file datasets/faces_preprocessed.pkl #3

Closed johnConsidine closed 8 years ago

johnConsidine commented 8 years ago

why is this coming up? there is no faces_preprocessed class in the folders of codes

mbeyeler commented 8 years ago

Have a look at the top of chapter7.py, the process flow is explained there. After you recorded your own training set, which is stored in datasets/faces_training.pkl, you want to train the MLP on it by running train_test_mlp.py. This script will create datasets/faces_preprocessed.pkl.

Chapter 7: Learning to Recognize Emotion in Faces
An app that combines both face detection and face recognition, with a
focus on recognizing emotional expressions in the detected faces.
The process flow is as follows:
 * Run the GUI in Training Mode to assemble a training set. Upon exiting
   the app will dump all assembled training samples to a pickle file
   "datasets/faces_training.pkl".
 * Run the script train_test_mlp.py to train a MLP classifier on the
   dataset. This file will store the parameters of the trained MLP in
   a file "params/mlp.xml" and dump the preprocessed dataset to a
   pickle file "datasets/faces_preprocessed.pkl".
 * Run the GUI in Testing Mode to apply the pre-trained MLP classifier
   to the live stream of the webcam.
johnConsidine commented 8 years ago

so your saying mlp will create it? thanks for the response btw

mbeyeler commented 8 years ago

Exactly! You're welcome.

johnConsidine commented 8 years ago

you are telling me to run in "training mode" and "testing mode", im sorry if this is a dumb question but i just see options to "run" on spyder, i dont get these modes...

johnConsidine commented 8 years ago

also, where is datasets/faces_training.pkl??? the only files in datasets are "init" and "homebrew"

mbeyeler commented 8 years ago

Hi, as mentioned above, running chapter7.py in training mode will create datasets/faces_training.pkl. Running train_test_mlp.py will convert datasets/faces_training.pkl to datasets/training_preprocessed.pkl.

Regarding the GUI issue, you should see "Train" and "Test" buttons at the bottom: b04521_07_03

Click on "Train", select an emotion such as "happy", then smile into your webcam and click "Take Snapshot". Keep taking snapshots of all emotions so that you'll get a good training set (I'd say, at least 10 snapshots per emotion). Then when you close the GUI, all snapshots will be dumped to datasets/faces_training.pkl.

However, there seems to be a bug in the Windows version of wxPython (see issue #4), which messes up the GUI. I'm working on it and will post an update as soon as I have found a workaround.

johnConsidine commented 8 years ago

ok thanks for the help :)

G10DRAS commented 8 years ago

Hello Michael, Is it possible to train MLP (params/mlp.xml) using multiple datasets/faces_training-*.pkl files? Basically I want to do continuous Trainingand Testing in multiple sessions.

mbeyeler commented 8 years ago

Hi guys, I just created a Google group that would be better suited for discussions. Issues are more for bug reports... Maybe we can discuss options over there?

G10DRAS commented 8 years ago

agree.

johnConsidine commented 8 years ago

ok just let me know if you get the gui working

johnConsidine commented 8 years ago

it keeps freezing when i try to run it... when it runs only a picture of me shows up and where the buttons should be its not coming on

johnConsidine commented 8 years ago

ill email you

mbeyeler commented 8 years ago

Reportedly, this issue has been resolved. Future readers/users, you are welcome to discuss issues and seek help in our brand-new Google group.

XDUSPONGE commented 8 years ago

Hello Michael Is this code suitable for linux , i just run the code in ubuntu and ,the test button can not be clicked, it is very strange. It just says that there should be at least one input

mbeyeler commented 8 years ago

Hi XDUSPONGE, I know you're aware of the answer by now, but I'm posting this here for the sake of others: This issue has been discussed in the Google Discussion group.

In short, you need to generate a training set first by following the workflow I mentioned in the second post above from Feb16. This excerpt is taken from the source code of chapter7/chapter7.py. A more detailed explanation and step-by-step guide can be found in the book in Chapter 7.

If you have more questions or run into any issues with this, it's easier to ask in the Discussion group. Thanks!