Pedestrians Instance Segmentation with Mask R-CNN using Pytorch

Pedestrians detection and segmentation with Mask R-CNN using transfer learning on PennFudanPed Dataset Implemented in Pytorch

Dataset

PennFundanPed Dataset consist of 170 images of pedestrians and segmentation mask for each image

Dataset Link: (https://www.cis.upenn.edu/~jshi/ped_html/)

Neural Network

• Used a pretrained Mask R-CNN

• Transfer learning on Faster R-CNN used by Mask R-CNN to modify the fully connected layers responsiple for classfication and bounding boxes regression

• Transfer learning on last layer of the mask detection of Mask R-CNN

• Fixed feature extractor

Examples

Semantic Segmentation (using same color for each class)