Problems in training data/test data

Hi, I have an issue with the training data. cvs_to_numpy.py outputs ‘labels_kike.npy’ and ‘data_kike.npy’, where the emotion_recognition.py searches for ‘data_set_fer2013.npy’ and ‘data_labels_fer2013.npy’ while training and cannot find it. Also using poc, I guess it cannot find the image instead of the faces. I get the error message:

AttributeError: 'NoneType' object has no attribute 'shape'

Can you help with these?

[sakshamjindal07]

Hi @ghulamsiddiq Do this. Copy constants.py to the data folder. In the copied file, change the CASC_PATH to CASC_PATH = ‘haarcascade_frontalface_default.xml’. Now copy the haarcascade files to the ‘data’ folder which is your working folder presently where cvs_to_numpy.py is placed. This works.

[RobinManhas]

Steps i took to resolve the issues in training:

1. Downloaded training set of images from Kaggle

Link: https://www.kaggle.com/c/challenges-in-representation-learning-facial-expression-recognition-challenge/data

1. Run the cvs_to_numpy.py script. If you encounter an error like “slicing indices must be Integer or None”, this is because from Python 3.x onwards, int/int gives a float. Just change line 16 to:

gray_border[((150 // 2) - (SIZE_FACE//2))😦(150//2)+(SIZE_FACE//2)), ((150//2)-(SIZE_FACE//2))😦(150//2)+(SIZE_FACE//2))] = image

1. Once the script ends successfully, 2 files with the name “data_kike.npy” and “labels_kike.npy” will be created. I renamed the “data_kike.npy” to “data_set_fer2013.npy” and “test_set_fer2013.npy”. (Not correct to have same data and test set, but wanted to sanity check the working of scripts). Same for labels. Changing the names and adding these 4 files: data_set_fer2013.npy, test_set_fer2013.npy, data_labels_fer2013.npy, test_labels_fer2013.npy, in the /data/… folder fixes the issues in training data.