This project was my internship project that was about detecting smile, laugh and poker faces. I have used torch library and resnet34 pretrained model for training my model. You can find the main dataset which contains not-resized and non-cropped images hereunder.
main.py contains main code for train the dataset. The dataset contains cropped faces from a main dataset that have croped with face_detection.py code. The dataset have 3 classes: laugh, poker, and smile. Laugh and smile folders have about 900 images and poker folder has 373 images. Change your path for training the model. The resize.py is a code for resize all images in a folder and the facerecognition.py is a code for cropping faces from all images.
After running the main.py you have a (my_resnet34_lr0.03_SGD_model.pth) file and you will use it in Predict-image.py code for predicting your images. The output of predict-image.py is like the picture below.
