Bazingaa Team , Problem Statement :Unacademy
Bazingaa Team , Problem Statement :Unacademy
Unacademy -Identify and Remove Background from a video stream Team Name: Bazingaa
•Steps:1.Convert video stream to frames.
2.Apply SemanticSegmentation: compute a pixel-wise mask for every object in the image. Modelsbeing considered: a.MaskR-CNN b.UNET
3.Filter out all other objects except ‘person’in the frame.
4.Add the frame to a video writer.
•MaskR-CNN (Reference:https://github.com/matterport/Mask_RCNN)
-Mask R-CNN process: -After semanticsegmentation(test frame1): -After background removal(test frame1):
•UNET:The U-Net architecture is built upon the Fully Convolutional Network and modified in a way that it yields better segmentation. Reference: https://github.com/tensorflow/tfjs-models/tree/master/mobilenet
After background removal using MobileNet API: •User Interface: A simpleweb pageis created usingJavaScript, HTML. •Highlights: o Pretrained frozen TensorFlow model is being used & trained by Tensor Core through portrait datasets from Flickr. o The TensorFlow JS Layer Model with quantization level is being used to make it lightweight(approximately 2 MB). o Front-end being developed allows live manipulation of the video from Integrated Webcam, IP Camera, orStreaming Videos.
Result 1:
Result 2:
Result 3:
