Instance Segmentation Using Mask R-CNN on Custom Dataset

Implementation of Mask RCNN on Custom dataset.

Github: https://github.com/AarohiSingla/Mask-R-CNN-using-Tensorflow2

Explained:
1- How to annotate the images for Mask Rcnn?
2- Train Mask RCNN model
3- Test Mask Rcnn model.
4- Calculate mean average precision mAP
5- Confusion Matrix plotted

For any queries: [email protected]


Feature Extractor: 
The first step is to take an image and extract features using the ResNet 101 architecture. These features act as an input for the next layer which is Region Proposal Network.
Region Proposal Network RPN:
We take the feature maps obtained in the previous step and apply a region proposal network .This basically predicts if an object is present in that region or not. In this step, we get those regions which the model predicts contain some object. RPN does that using a binary classifier on a lot of boxes over the image and returns object/no-object scores. Anchors with high objectness score are passed to the stage two to be classified.
This will return the candidate bounding boxes.
Region of Interest RoI:
The regions obtained from the RPN might be of different shapes, right? Hence, we apply a pooling layer and convert all the regions to the same shape.

Object detection: The second branch of the network takes the RoI feature maps and predicts class labels and bounding box offsets for each proposal. This is similar to the classification and regression branches in Faster R-CNN.

Mask prediction:
Once we have the RoIs based on the IoU values, we can add a mask for each region that contains an object. 
This third branch of the network takes the RoI feature maps and generates a binary mask for each region proposal. This is achieved using a fully convolutional network that generates a pixel-wise mask for each object. where each pixel is assigned a value of 1 if it belongs to the object and 0 otherwise. This means that for each object instance in the image, Mask R-CNN produces a separate binary mask that indicates the exact location and shape of the object.


#imagesegmentation #maskrcnn #objectdetection #python #tensorflow