MUHAMMAD AHMED RAZA
20021519-050
SYED HASEEB
20021519-058
OBJECT DETECTION & OBJECT
SEGMENTATION
Object Detection is a
fundamental task in
computer vision locating
Object
Object Segmentation a step
further it provides
Medical
boundaries at the pixel
level
Application in Various
Autonomous Vehicles
Industries
Industries
(manufacturing ,robotics)
LIMITATIONS OF
TRADITIONAL METHODS
• Computational Complexity (R-CNN,Fast R-CNN)
• Difficulty handling Overlapping Instances
• Limited Spatial Precision (location , position)
Must Need A New Technology!
These limitations show the need of new technique
which is Mask R-CNN
EVOLUTION FROM FASTER R-CNN
TO MASK R-CNN
Mask R-CNN
Faster R-CNN
build upon
does not address
Faster R-CNN
the pixel level
by adding Mask-
segmentation
Head
ROI Align
Mask-Head is
ensure more
responsible for
accurate pixel-
pixel level
to-pixel
segmentation
alignment
ARCHITECTURE OF MASK R-CNN
Backbone Network (Extract features
from input image)
Region Proposal Network RPN
(Main Purpose Bounding
Boxes,followed byNMs to filter
redundant boxes)
ROI Align (old method use fixed grid,
it use bilinear interpolation,flexible )
ROI Classification and Bounding Box
Regression (undergo for object
labeling)
Instance Segmentation (Detailed
mask for each proposed regions)
LOSS FUNCTION
Classification Bounding Box
RPN loss Mask Loss
Loss Regression Loss
