Detection in video

[kaimonhtun]

How to test detection for video file?

[duanzhiihao]

Could you provide more information?

If you mean testing with no metric evaluation, you can just test on every single frame like in

RAPiD/api.py

Line 51 in e56ac87

def detect_one(self, **kwargs):

If you mean testing with metric evaluation, I guess you need to write your own script to do this.

[seino-tsuyoshi]

For example,

from api import Detector
import numpy as np
import cv2

from PIL import Image
def cv2pil(imgCV):
    imgCV_RGB = cv2.cvtColor(imgCV, cv2.COLOR_BGR2RGB)
    imgPIL = Image.fromarray(imgCV_RGB)
    return imgPIL  

detector = Detector(model_name='rapid',
                    weights_path='./weights/pL1_MWHB1024_Mar11_4000.ckpt')

cap = cv2.VideoCapture('videos/test.mp4')

while(cap.isOpened()):
    # フレームを取得
    ret, frame = cap.read()

    if ret == False:
        break

    np_img = detector.detect_one(pil_img=cv2pil(frame),
                    input_size=1024, conf_thres=0.3,
                    return_img=True)
    new_image = np.array(np_img, dtype=np.uint8)
    np_img = cv2.cvtColor(new_image, cv2.COLOR_RGB2BGR)
    # フレームを表示
    cv2.imshow("Frame", np_img)

    # qキーが押されたら途中終了
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

cap.release()
cv2.destroyAllWindows()

However, this code has duplicate value error in "pil_img" on api.py.
So you can modify the code below,

api.py #L73

RAPiD/api.py

Line 72 in fcb764a

detections = self._predict_pil(img, **kwargs)

to

detections = self._predict_pil(**kwargs)

and L116

RAPiD/api.py

Lines 116 to 130 in fcb764a

	def _predict_pil(self, pil_img, **kwargs):
	'''
	Args:
	pil_img: PIL.Image.Image
	input_size: int, input resolution
	conf_thres: float, confidence threshold
	'''
	input_size = kwargs.get('input_size', self.input_size)
	conf_thres = kwargs.get('conf_thres', self.conf_thres)
	assert isinstance(pil_img, Image.Image), 'input must be a PIL.Image'
	assert input_size is not None, 'Please specify the input resolution'
	assert conf_thres is not None, 'Please specify the confidence threshold'
	# pad to square
	input_img, _, pad_info = utils.rect_to_square(pil_img, None, input_size, 0)

to

    def _predict_pil(self, **kwargs):
        '''
        Args:
            pil_img: PIL.Image.Image
            input_size: int, input resolution
            conf_thres: float, confidence threshold
        '''
        input_size = kwargs.get('input_size', self.input_size)
        conf_thres = kwargs.get('conf_thres', self.conf_thres)
        assert isinstance(kwargs.get('pil_img', None), Image.Image), 'input must be a PIL.Image'
        assert input_size is not None, 'Please specify the input resolution'
        assert conf_thres is not None, 'Please specify the confidence threshold'

        # pad to square
        input_img, _, pad_info = utils.rect_to_square(kwargs.get('pil_img', None), None, input_size, 0)

It works for me.

https://youtu.be/6h19o3Sj0UA