Please use this identifier to cite or link to this item:
http://hdl.handle.net/10609/146733
Title: | Face Recognition con imágenes de Data Streaming con modelos generados a partir de Transfer Learning y diferentes Funciones de Activación |
Author: | Ramirez Subeldia, Elvio |
Tutor: | Ortiz Santiago, Victor Alejandro |
Others: | Daradoumis Haralabus, Atanasi Florit Medina, Xavier Borja Matas, Jaime Borja |
Keywords: | face recognition transfer learning deep learning |
Issue Date: | 25-Jun-2022 |
Publisher: | Universitat Oberta de Catalunya (UOC) |
Abstract: | A challenge when you want to work training neuronal artificial network, is to create a specific architecture and initial values that can start your neural network. This investigation work, using Transfer Learning and different Activation Functions to create a model for Face Recognition with Video Data Streaming, raises a result of comparative between different models. To solve this analysis, three modules are created, the first one dataset creator module, second training module, third Face Recognition module. In dataset creator module we take Transfer Learning from OpenCV. We will use the library for face detection ‘Haar Cascades’ and capturing images from webcam will be possible to create the dataset. Training module, in this module we used Transfer Learning from VGG16 library, facilitating the creation of the corresponding architecture for a convolutional network. Also, we implemented ‘imagenNET’ to get the initial values for the network. Adapting each neuronal network for each Activation Function ‘ReLu’, ‘Swish’ and ‘Mish’. Keras and Tensorflow library will facilitate the use of the different network layers, getting a trained model for Face Recognition with metrics of cost and functioning. Finally, in Face Recognition module, we import the result of the model trained and Face recognition is tested. Analyzed the results of the tests, no conclusive results are obtained on which Activation Function (‘ReLu’, ‘Swish’, ‘Mish’) gives the best result for Face Recognition, but the model remains for future lines of research where the hyperparameters of the neural network can be tuned. |
Language: | Spanish |
URI: | http://hdl.handle.net/10609/146733 |
Appears in Collections: | Bachelor thesis, research projects, etc. |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
eramirez225TFM0622memoria.pdf | Memoria del TFM | 1,69 MB | Adobe PDF | ![]() View/Open |
Share:


This item is licensed under aCreative Commons License