SSS-Vision is a web-based, flask powered AI system that could receive an image of tripleS members from the user and return a face-recognized and labeled image. The model uses MTCNN and Facenet pretrained model to recognize the face, and i'm using a python-based web-scraper to get the data (pictures of tripleS members) from Pinterest.
opencv2 FaceLBPHRecognizer (haarcascade_frontalface.xml) using only 2 members, S1 SeoYeon and S7 NaKyoung turned out great but it struggles with side-profiles so it is replaced with mtcnn for the face recognition
- flask -> for the web server (request and send_file is like the name suggest)
- mtcnn -> to perform face detection in images
- cv2 -> image reading, recoloring, and resizing for training purposes
- numpy -> to save the embedded faces, numpy array manipulation, etc
- torch -> to manipulate the facial region of the image into tensors to prepare for training
- facenet/inceptionresnetv1 -> the base pretrained model to compare the faces
- cosine_similarity -> imported from sklearn.metrics.pairwise to calculate the similarity of the tensored face and the stored face embeddings
- Image -> imported from PIL for impace processing to be sent from the model to the front-end
- io -> same as Image
- os -> to go into every folder and join directories for training purposes
- glob -> honestly damn near useless, its to just find out how many images are in the folder
- Prototyping = https://youtu.be/oXlwWbU8l2o?si=XJi3nVT9YhfxdC38&t=10146
- Flask = https://youtu.be/Z1RJmh_OqeA?si=BJCojDmc5RgcpbGR
- HTML CSS = Basic is from brocode, and past projects with BNCC
- JavaScript = AJAX tutorial from BNCC, and ChatGPT
- Neural Network stuff = https://youtu.be/-rrxxpiZa00?si=h5BGQWerr0EDCQ31