What is a Mask R-CNN?

Mask R-CNN (Region-based Convolutional Neural Network) is a deep learning algorithm that is used for object detection and instance segmentation. Object detection is the task of identifying and locating objects within an image, while instance segmentation is the task of identifying and segmenting each object instance within an image.
Mask R-CNN is an extension of Faster R-CNN, which is a popular object detection algorithm. Faster R-CNN replaces the traditional sliding window method with a region proposal network (RPN), which generates region proposals instead of windows. This makes the algorithm faster and more accurate than traditional methods.
In Mask R-CNN, an additional branch is added to the Faster R-CNN architecture to predict the object mask. This branch is called the "mask branch", and it takes the features from the last convolutional layer of the CNN as input, and outputs a binary mask for each object instance. The mask branch is trained to segment the object instances in the image, allowing for more accurate object instance segmentation and object detection.
Additionally, The term "pretrained" in this context refers to pre-trained model on a dataset that have been trained on a task that is similar to the one you need to use them. This provide a good starting point for fine tuning the model.