Abstract: The project focuses on leveraging OpenCV's face detection algorithms to identify and locate human faces in a dataset. The validation dataset is used to verify the results, and the performance is reported on the test data.

1. Introduction:

Face detection, a critical aspect of computer vision, involves automatically locating human faces in visual media. The project employs OpenCV, a powerful computer vision library, to implement face detection. The detected faces are reported with their positions, sizes, and orientations.

2. Brief Overview of Face Detection in OpenCV:

OpenCV initially used "Haar Cascade face detection." However, with the evolution of deep learning, newer versions include deep learning face detection models. The project utilizes a Single Shot Face Detector based on the ResNet architecture. This model is originally from Caffe and is integrated into OpenCV. The implementation supports models in TensorFlow, Caffe, and PyTorch/Torch.

3. Prerequisites for Caffe OpenCV Face Detection:

Caffe models, contributed by rybnikov, are essential for this project. The Caffe-based face detector resides in the face detector sub-directory of the DNN samples. When working with OpenCV's deep neural network module and Caffe models, two sets of files are required: the .prototxt file(s) defining the model architecture and the .caffemodel file containing the layer weights.

4. Results:

The project generated results on the validation dataset and computed the F1 score using Compute_Fbeta.py. The face detection algorithm provides the coordinate boundaries of the faces, along with confidence scores. To minimize false positives, a confidence threshold of 90% is selected. The achieved F1 score for face detection is 0.85, indicating robust performance.