Scribd
Upload
23 views1 page

Supriya - Naik - Working Process of Video Based Emotion Detection

The document discusses a working process for video-based emotion detection using a dataset containing 303 video clips with angry and neutral emotions. Various CNN models were tested and optimized using techniques like pre-trained Resnet50 to achieve 86% accuracy in classifying emotions, higher than the base paper.

Uploaded by

karandesaurabh0901
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
23 views1 page

Supriya - Naik - Working Process of Video Based Emotion Detection

The document discusses a working process for video-based emotion detection using a dataset containing 303 video clips with angry and neutral emotions. Various CNN models were tested and optimized using techniques like pre-trained Resnet50 to achieve 86% accuracy in classifying emotions, higher than the base paper.

Uploaded by

karandesaurabh0901
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 1

Working process of Video based Emotion detection

1. ISED dataset is used for detecting emotions. In which 2 emotions are


been added i.e angry and netural.

2. Before 428 video clips where their in a dataset but after balancing and
adding 2 emotions in a dataset total 303 video clips are persent.

3. Training of model is been done.

4. At first 3 different model is being used RNN, LSTM, CNN comparison is


been done of this 3 model from which CNN is the best model for
emotion detection.

5. To increase the accuracy of model manipulations of layers in CNN is


been done.

6. Comparison is been done of all optimizer in order to increase the


accuracy, from which Adam optimizer gives best results so it been used.

7. Best activation function for multi-class classifier softmax is been used.

8. First tried to increase the accuracy by increasing the layers in CNN


model, but it did not increased more than 70%.

9. So tired to use pre-trained model Resnet50 with CNN architecture so


to increase accuracy. Accuracy is 86% which is more than the base
paper.

10. Confusion matrix is also been plotted for 6 different emotions.

You might also like