src/tracker/signboard_segment/main.py

from signboard_detect import inference_signboard
import os
import argparse
import tqdm
import cv2
import numpy as np
from PIL import Image

def compose(output, mask):
    h,w = mask.shape

    for i in range(0, h):
        for j in range(0,w):
            if (mask[i,j] > 0.5):
                output[i,j] = 255
    return output

def get_parser():
    parser = argparse.ArgumentParser(description="Signboard Detection")

    parser.add_argument("--input",
                        type=str,
                        default="./images",
                        help="A list of space separated input images")
    parser.add_argument("--output",
                        type=str,
                        default="./output/output_signboard",
                        help="A list of array of segmentation")
    parser.add_argument("--checkpoint",
                        type=str,
                        default="./checkpoints/ss/ss.ckpt",
                        help="File path to best model checkpoint")

    args = parser.parse_args()
    return args


def handle(args):
    if args.input:
        if os.path.isdir(args.input):
            args.input = [os.path.join(args.input, fname)
                          for fname in os.listdir(args.input)]
        elif os.path.isfile(args.input):
            args.input = [args.input]

    for path in tqdm.tqdm(args.input):
        print(path)
        img = cv2.imread(path)
        
        image = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
        im_pil = Image.fromarray(image)
        
        dimensions = img.shape
        hei, wid = dimensions[0], dimensions[1]
        print(hei, wid)
        result = inference_signboard(im_pil, args.checkpoint)
        print(" **************** Result **************** ")
        print(result['rois'].shape)
        print(result['masks'].shape)
        print(result['class_ids'].shape)
        print(result['scores'].shape)
        print(" **************************************** ")
        for box in result['rois']:
            box = box.tolist()
            image = cv2.rectangle(img, (int(box[0]), int(
                box[1])), (int(box[2]), int(box[3])), (255, 0, 0), 2)
        root_ext = os.path.splitext(path)
        output_path = os.path.join(args.output, root_ext[0] + "_output" + root_ext[1])
        cv2.imwrite(output_path, image)
        
        img_output = np.zeros((hei,wid), dtype="uint8")
        for j in range(0,len(result['masks'])):
            mask = result['masks'][j]
            im_np = np.array(mask)
            img_output = compose(img_output, im_np)
        output_path = os.path.join(args.output, root_ext[0] + "_mask" + root_ext[1])
        cv2.imwrite(output_path, img_output)


def main():
    args = get_parser()
    handle(args)


if __name__ == "__main__":
    main()