This is a Pytorch implementation of retraining a person segmentation model, specifically the MaskRCNN Model provided by Torchvision. This reposity provides two folders:
train/: This contains the training script, that can be run on Google Colab and contains cells for downloading.api/: This provides an API to perform inference on your image.
Note : Refer to respective README.md inside these folders for a more detailed description.
MaskRCNN is based on FasterRCNN model.
Mask R-CNN adds an extra branch into Faster R-CNN, which also predicts segmentation masks for each instance.
The backbone used in the model is resnet50. The model was optimized for providing the highest IoU on the validation set.
The model can be downloaded here.