/BasicsOf_ANN_CourseWork-GRIAT20-21-

Primary LanguageJupyter Notebook

BasicsOf_ANN_CourseWork-GRIAT20-21

Roadsign detection in DVR video

This repo contains .ipymb files with trained models

  1. YOLOV4
  2. Faster-RCNN
  3. RetinaNet-101

For Faster-RCNN and RetinaNet, Facebook's Detectron2 platform was used.

  • Initial Dataset is in Yolo format.
  • For Faster-RCNn and RetinaNet, dataset annotations are converted to its corresponding one using RoboFlow
  • Outputs are inside the foldes. Video detection is done only to yolov4, for others are in process

Dataset

  • Dataset is initially in Yolo format
  • Dataset divided into 518 train,148 valid, 74 test sets
  • There are 4 classes in dataset
  • For Faster-RCNN and RetinaNet, dataset is converted to its corresponding annotation