How Instrumental Finance complements the Composable ecosystem

As Instrumental Finance works towards creating new cross-layer yield generation opportunities, it aligns itself with Composable Finance, the face of DeFi’s future. Composable is growing in its…

独家优惠奖金 100% 高达 1 BTC + 180 免费旋转

Facial Expressions Recognition using Keras

Testing the model with a live webcam using OpenCV

The objective of the project:- The aim of this project is to identify people’s emotions based on their live video by building and training a system that automatically monitors people's emotions and expressions with more than 24000 facial images, with their associated facial expression labels. The chosen labels were angry, happy, neutral, sad, and surprise.

Part 1: Importing the libraries and loading the dataset

Part 2: Data augmentation to increase the amount of data artificially by generating new data points from existing data.

Part 3: We will create a sequential deep learning model based on Convolutional Neural Network and Residual Blocks to predict the facial expression.

Part 4: Optimising the training using ModelCheckpoint, EarlyStopping, ReduceLROnPlateau, and compiling the model.

Part 5: Test the model using OpenCV

Both these operations are performed on both the training and validation datasets.

Among 4 deep neural networks, one of the neural networks is shown above. It includes conv2D block, activation layer, batch normalization, and max-pooling.

Now we want the weighted sum of the input to be transformed into an output from the nodes in a layer of the network, thus we use the ‘elu’ activation function, ELU is very similar to RELU except for negative inputs. They are both in identity function form for non-negative inputs. On the other hand, ELU becomes smooth slowly until its output is equal to −α whereas RELU sharply smoothes.

Batch normalization is a layer that allows every layer of the network to do learning more independently. It is used to normalize the output of the previous layers and often we add it after the activation layer for better results.

Maximum pooling is done which calculates the maximum, or largest, value in each patch of each feature map. The results are downsampled or pooled feature maps that highlight the most present/important feature in the patch.

Now a dropout layer with 20% is added to prevent neural networks from Overfitting. Dropout is a technique where randomly selected neurons are ignored during training.

After 4 layers flatten and dense layers are added together with ‘elu’ activation, batch normalization, and dropout layer where:

After this 3 dense layers are added to a fully connected layer and the SOFTMAX activation function which is normally applied to the very last in the neural network. Softmax is useful because it converts the output of the last layer in the neural network into a probability distribution. It is generally used as the activation function for multi-class classification problems where class membership is required on more than two class labels here also we need 5 class labels.

ModelCheckpoint callback -

EarlyStopping callback-

ReduceLROnPlateau callback-

COMPILING THE MODEL- It includes loss function as ‘categorical crossentropy’, optimizer as ADAM, and accuracy metrics.

Lastly, we fit our training model. Hyperparameters used are self-explanatory.

Part 5.1: Importing the libraries- Keras, OpenCV as cv2, numpy.

Part 5.2: Introducing the Cascade classifier- Which is an Object Detection Algorithm used to identify faces in an image or a real-time video. The algorithm uses edge or line detection features. Loading the model saved in the.h5 file and defining the class labels as ‘Angry’, ‘Happy’, ‘Sad’, ‘Neutral’, ‘Surprise’.

Part 5.3: Testing using the inbuilt webcam- VideoCapture(0) turns on the laptop’s inbuilt camera.