Opening BDD dataset and saving it as a masked image

bdd dataset tool

import json
import numpy as np
import os 
import skimage
import cv2
import time 
import scipy.misc
from PIL import Image
dataset_dir = "/media/ixtiyor/New Volume1/datasets/bdd/bdd100k_images/bdd100k/images/100k/"
input_dir = 'val'
annotations = json.load(open(os.path.join(os.path.join(dataset_dir, input_dir), "via_region_data.json")))
annotation_root = os.path.join(os.path.join(dataset_dir, "annotation"), input_dir)
def get_mask_from_polygons(polygon, height, width):
        mask = np.zeros([height, width,3],
                        dtype=np.uint8)
        #print ('mask shape',mask.shape)
        for i, p in enumerate(polygon):
#             print ('i',i)
#             print ('p',p)
            # Get indexes of pixels inside the polygon and set them to 1
            x=[]
            y=[]
            for d in p:
                for l in d['vertices']:  
                    x.append(l[0])
                    y.append(l[1]-2)
#             print ('x',x)
#             print ('y',y)
            rr, cc = skimage.draw.polygon(y,x)
            print(x, y)
            print(rr,cc)
            mask[rr, cc] = [255,255,255]
#         print(mask)
        return mask
i = 0
categories = []
for a in annotations:
    i += 1
    start = time.time()
    
    categories.extend([d['category'] for d in a['labels']])
    # print(np.unique(categories))
    polygons=[d['poly2d'] for d in a['labels'] if d['category']=="lane"]
    # print(polygons)
    image_path = os.path.join(dataset_dir, a['name'])
    height, width = [720, 1280]#image.shape[:2]
    # self.images.append(image_path)
    img = get_mask_from_polygons(polygons,height, width) 
    # print(np.unique(img))
    img = np.array(img)
    
    # print(np.unique(img))
    mask_path = os.path.join('/media/ixtiyor/087E6CE67E6CCDCE/annotation/',input_dir, a['name'].split('.')[0] +".png" )
    # img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
    # cv2.imwrite(mask_path, img, [cv2.IMWRITE_JPEG_QUALITY, 255])
    # scipy.misc.toimage(img, cmin=0.0, cmax=255.0).save(mask_path)
    # im = Image.fromarray(img,"RGB" )
    # im.save(mask_path)
    # read_image = cv2.imread(mask_path, cv2.IMREAD_UNCHANGED)
    # print(np.shape(img), type(img))
    # np.save(mask_path,img)
    # read_image =  np.load(mask_path)
    # print()
    # print(i, "qolgan vaqt " , (time.time()-start) * (10000-i) * 1./60, " min")
    # break
    cv2.imshow("img", img)
    if(cv2.waitKey(3) >0):break