detection.py

from detectron2.engine import DefaultPredictor
import detectron2
from detectron2.utils.logger import setup_logger
setup_logger()
from detectron2.utils.video_visualizer import VideoVisualizer
from detectron2.config import get_cfg
from detectron2.data import MetadataCatalog
from detectron2.utils.visualizer import ColorMode, Visualizer
from detectron2 import model_zoo
from detectron2.data.datasets import register_coco_instances
from PIL import Image 
import PIL 
import cv2
import numpy as np
import matplotlib.pyplot as plt


class Detector:

    def __init__(self, model_type = "object_detection"):
        self.cfg=get_cfg()
        self.cfg.merge_from_file(model_zoo.get_config_file("COCO-Detection/faster_rcnn_R_50_FPN_1x.yaml")) # load  the default configuration
        self.cfg.MODEL.WEIGHTS = 'model_final.pth'
        self.cfg.MODEL.ROI_HEADS.SCORE_THRESH_TEST = 0.8
        self.cfg.MODEL.ROI_HEADS.NUM_CLASSES = 2
        self.cfg.MODEL.DEVICE="cpu"

        dataset_name="guns"
        classes=['guns','Gun']
        MetadataCatalog.get(dataset_name).set(thing_classes=classes)

        self.predictor = DefaultPredictor(self.cfg)

    def onImage(self, imagePath):
        image = cv2.imread(imagePath)
        predictions = self.predictor(image)
        dataset_name="guns"

        viz = Visualizer(image,MetadataCatalog.get(dataset_name),scale=1)
        
        output = viz.draw_instance_predictions(predictions['instances'].to('cpu'))
        filename = 'result.jpg'
        cv2.imwrite(filename, output.get_image()[:,:,::-1])
        # cv2.waitKey(0)
        # cv2.destroyAllWindows()