No KITTI formatted data is generated

[SIGMIND]

The data2kitti.py file not generating the KITTI formatted data in the output folder.

[ak-nv]

@SIGMIND May I know; if you get any error? Or What is your log?

[SIGMIND]

I only wanted to generate KITTI formatted label files from WIDER Face dataset. Heres the modified data2kitty.py

from data_utils.widerface2kitti import widerFace2kitti
from data_utils.mafa2kitti import mafa2kitti
from data_utils.fddb2kitti import fddb2kitti
from data_utils.kaggle2kitti import kaggle2kitti
from data_utils.argparser_data2kitti import argparser_data2kitti
from data_utils.check_labels import test_labels
import os

def main():
    # Set Parameters
    arg_parser = argparser_data2kitti()
    args = arg_parser.make_args()
    widerface_base_dir = r'./WIDER-Face-Dataset' #args.widerface_dataset_path # Use only from selected sub-folders
    # Store Converted annotations in KITTI format
    kitti_base_dir = r'./KITTI-WIDER' #args.kitti_base_path
    category_limit = [args.category_limit, args.category_limit]  # Mask / No-Mask Limits
    kitti_resize_dims = (args.tlt_input_dims_width, args.tlt_input_dims_height)  # Default for DetectNet-v2 : Look at TLT model requirements

    total_masks, total_no_masks = 0, 0
    count_masks, count_no_masks = 0, 0
    # To check if labels are converted in right format
    if args.check_labels:
        # Check from train directory
        test_labels(kitti_base_dir=kitti_base_dir + '/train/', file_name=args.label_filename)
    else:

        # ----------------------------------------
        # Wider-Face Dataset Conversion
        # --------------------------------------
        total_masks += count_masks
        total_no_masks += count_no_masks
        print("Total Mask Labelled:{} and No-Mask Labelled:{}".format(total_masks, total_no_masks))
        category_limit_mod = [category_limit[0] - total_masks, category_limit[1] - total_no_masks]

        if args.train:
            annotation_file = os.path.join(widerface_base_dir, 'wider_face_split/wider_face_train.mat')
            widerFace_base_dir = os.path.join(widerface_base_dir, 'WIDER_train/images')
        if args.val:
            # Modify this
            annotation_file = os.path.join(widerface_base_dir, 'wider_face_split/wider_face_val.mat')
            widerFace_base_dir = os.path.join(widerface_base_dir, 'WIDER_val/images')

        kitti_label = widerFace2kitti(annotation_file=annotation_file, widerFace_base_dir=widerFace_base_dir,
                                      kitti_base_dir=kitti_base_dir, kitti_resize_dims=kitti_resize_dims,
                                      category_limit=category_limit_mod, train=args.train)
        count_masks, count_no_masks = kitti_label.mat2data()
        total_masks += count_masks
        total_no_masks += count_no_masks
        print("----------------------------")
        print("Final: Total Mask Labelled:{}\nTotal No-Mask Labelled:{}".format(total_masks, total_no_masks))
        print("----------------------------")

if __name__ == '__main__':
    main()`

But executing this only outputs total counts, but no files in KITTI-WIDER folder.

root@b098736d6bc2:/workspace/face-mask-detection# python3 data2kitti.py --widerface-dataset-path WIDER-Face-Dataset --train
Total Mask Labelled:0 and No-Mask Labelled:0
WideFace: Total Mask Labelled:0 and No-Mask Labelled:6007
----------------------------
Final: Total Mask Labelled:0
Total No-Mask Labelled:6007
----------------------------
root@b098736d6bc2:/workspace/face-mask-detection# python3 data2kitti.py --widerface-dataset-path WIDER-Face-Dataset --train
Total Mask Labelled:0 and No-Mask Labelled:0
WideFace: Total Mask Labelled:0 and No-Mask Labelled:6007
----------------------------
Final: Total Mask Labelled:0
Total No-Mask Labelled:6007
----------------------------

[ak-nv]

If you just want to generate for WiderFace Dataset; you can do python3 data_utils/widerface2kitti.py executing this script stand alone; change directories and add more to pick_list.

Main.py is currently modeled for 4K dataset only.

[SIGMIND]

@ak-nv yes, I have tried that too. But can't see the line which writes the label.txt in the corresponding folder.

[SIGMIND]

I have modified the widerface2kitty.py to capture all the faces (including occlusion)

import scipy.io
import os
from PIL import Image, ImageDraw

class widerFace2kitti():
    def __init__(self, annotation_file, widerFace_base_dir, kitti_base_dir, kitti_resize_dims, category_limit, train):
        self.annotation_file = annotation_file
        self.data = scipy.io.loadmat(self.annotation_file)
        self.file_names = self.data.get('file_list') # File Name
        self.event_list = self.data.get('event_list') # Folder Name
        self.bbox_list = self.data.get('face_bbx_list') # Bounding Boxes
        self.label_list = self.data.get('occlusion_label_list')
        self.kitti_base_dir = kitti_base_dir
        self.widerFace_base_dir = widerFace_base_dir
        self.count_mask = category_limit[0]
        self.count_no_mask = category_limit[1]
        self.kitti_resize_dims = kitti_resize_dims
        self.train = train
        self.len_dataset = len(self.file_names)
        #if self.train:
            # os.makedirs(self.kitti_base_dir+'/train/images',mode=0o777)
        self.kitti_images = os.path.join(self.kitti_base_dir, 'train/images')
            # os.makedirs(self.kitti_base_dir+ '/train/labels',mode=0o777)
        self.kitti_labels = os.path.join(self.kitti_base_dir, 'train/labels')
        #else:
            # os.makedirs(self.kitti_base_dir+'/test/images',mode=0o777)
           # self.kitti_images = os.path.join(self.kitti_base_dir, 'test/images')
            # os.makedirs(self.kitti_base_dir+'/test/labels',mode=0o777)
           # self.kitti_labels = os.path.join(self.kitti_base_dir, 'test/labels')

    def make_labels(self, image_name, category_names, bboxes):
        # Process image
        file_image = os.path.splitext(image_name)[0].split('\\')[1]
        img = Image.open(os.path.join(self.widerFace_base_dir, image_name)).convert("RGB")
        resize_img = img.resize(self.kitti_resize_dims)
        resize_img.save(os.path.join(self.kitti_images, file_image+'.jpg'), 'JPEG')
        # Process labels
        with open(os.path.join(self.kitti_labels, file_image+ '.txt'), 'w') as label_file:
            for i in range (0, len(bboxes)):
                resized_bbox = self.resize_bbox(img=img, bbox=bboxes[i], dims=self.kitti_resize_dims)
                out_str = [category_names[i].replace(" ", "")
                           + ' ' + ' '.join(['0'] * 1)
                           + ' ' + ' '.join(['0'] * 2)
                           + ' ' + ' '.join([b for b in resized_bbox])
                           + ' ' + ' '.join(['0'] * 7)
                           + '\n']
                label_file.write(out_str[0])

    def resize_bbox(self, img, bbox, dims):
        img_w, img_h = img.size
        x_min, y_min, x_max, y_max = bbox
        ratio_w, ratio_h = img_w / dims[0], img_h / dims[1]
        new_bbox = [str(x_min / ratio_w), str(y_min / ratio_h), str(x_max / ratio_w), str(y_max / ratio_h)]
        return new_bbox

    def mat2data(self):
        count = 0
        _count_mask, _count_no_mask = 0,0
        #pick_list = ['19--Couple', '13--Interview', '16--Award_Ceremony','2--Demonstration', '22--Picnic']
        # Use following pick list for more image data
        pick_list = ['2--Demonstration', '4--Dancing', '5--Car_Accident', '15--Stock_Market', '23--Shoppers',
                     '27--Spa', '32--Worker_Laborer', '33--Running', '37--Soccer',
                     '47--Matador_Bullfighter','57--Angler', '51--Dresses', '46--Jockey',
                     '9--Press_Conference','16--Award_Ceremony', '17--Ceremony',
                     '20--Family_Group', '22--Picnic', '25--Soldier_Patrol', '31--Waiter_Waitress',
                     '49--Greeting', '38--Tennis', '43--Row_Boat', '29--Students_Schoolkids']
        for event_idx, event in enumerate(self.event_list):
            directory = event[0][0]
            if any(ele in directory for ele in pick_list):
                for im_idx, im in enumerate(self.file_names[event_idx][0]):
                    im_name = im[0][0]
                    read_im_file = os.path.join(directory, im_name+'.jpg')
                    face_bbx = self.bbox_list[event_idx][0][im_idx][0]
                    category_id = self.label_list[event_idx][0][im_idx][0]
                    #  print face_bbx.shape
                    bboxes = []
                    category_names = []
                   # if _count_no_mask < self.count_no_mask:
                    for i in range(face_bbx.shape[0]):
                          xmin = int(face_bbx[i][0])
                          ymin = int(face_bbx[i][1])
                          xmax = int(face_bbx[i][2]) + xmin
                          ymax = int(face_bbx[i][3]) + ymin
                            # Consider only Occlusion Free masks
                           # if category_id[i][0] ==0:
                          category_name = 'No-Mask'
                          bboxes.append((xmin, ymin, xmax, ymax))
                          category_names.append(category_name)
                          _count_no_mask += 1
                       # try:
                           # if bboxes:
                          print("Label made")
                          self.make_labels(image_name=read_im_file, category_names= category_names, bboxes=bboxes)

                       # except:
                        #    pass

        print("WideFace: Total Mask Labelled:{} and No-Mask Labelled:{}".format(_count_mask, _count_no_mask))
        return _count_mask, _count_no_mask

    def test_labels(self, file_name):
        img = Image.open(os.path.join(self.kitti_images, file_name + '.jpg'))
        text_file = open(os .path.join(self.kitti_labels, file_name + '.txt'), 'r')
        features = []
        bbox = []
        category = []
        for line in text_file:
            features = line.split()
            bbox.append([float(features[4]), float(features[5]), float(features[6]), float(features[7])])
            category.append(features[0])
        print("Bounding Box", bbox)
        print("Category:", category)
        i=0
        for bb in bbox:
            draw_img = ImageDraw.Draw(img)
            shape = ((bb[0], bb[1]), (bb[2], bb[3]))
            if category[i] == 'No-Mask':
                outline_clr = "red"
            elif category[i] == 'Mask':
                outline_clr = "green"
            draw_img.rectangle(shape, fill=None, outline=outline_clr, width=4)
            i+=1

        img.show()

def main():
    widerFace_base_dir = r'WIDER-Face-Dataset' # Update According to dataset location
    kitti_base_dir = r'KITTI-WIDER' # Update According to KITTI output dataset location
    train = True # For generating validation dataset; select False
    #if train:
    annotation_file = os.path.join(widerFace_base_dir, 'wider_face_split/wider_face_train.mat')
    widerFace_base_dir = os.path.join(widerFace_base_dir, 'WIDER_train/images')
   # else:
        # Modify this
     #   annotation_file = os.path.join(widerFace_base_dir, 'MAFA-Label-Test/LabelTestAll.mat')
     #   widerFace_base_dir = os.path.join(widerFace_base_dir, 'test-images\images')

    category_limit = [10000, 10000] # Mask / No-Mask Limits
    kitti_resize_dims = (480, 272) # Look at TLT model requirements
    kitti_label = widerFace2kitti(annotation_file=annotation_file, widerFace_base_dir=widerFace_base_dir,
                             kitti_base_dir=kitti_base_dir, kitti_resize_dims=kitti_resize_dims,
                             category_limit=category_limit, train=train)
    count_masks, count_no_masks = kitti_label.mat2data()
  #  kitti_label.test_labels(file_name='0_Parade_Parade_0_371')
if __name__ == '__main__':
    main() 

But facing error:

root@318833e45ec6:/workspace/face-mask-detection# python widerface2kitti.py 
Label made
Traceback (most recent call last):
  File "widerface2kitti.py", line 146, in <module>
    main()
  File "widerface2kitti.py", line 143, in main
    count_masks, count_no_masks = kitti_label.mat2data()
  File "widerface2kitti.py", line 93, in mat2data
    self.make_labels(image_name=read_im_file, category_names= category_names, bboxes=bboxes)
  File "widerface2kitti.py", line 33, in make_labels
    file_image = os.path.splitext(image_name)[0].split('\\')[1]
IndexError: list index out of range

[Ziegan]

Whlie running the data2kitti.py by giving the required paths

I get output like this:

Total Mask Labelled:0 and No-Mask Labelled:0 Traceback (most recent call last): File "data2kitti.py", line 107, in main() File "data2kitti.py", line 58, in main kitti_label = mafa2kitti(annotation_file=annotation_file, mafa_base_dir=mafa_base_dir, UnboundLocalError: local variable 'annotation_file' referenced before assignment

Can anyone help me with this error?

[ak-nv]

May I know, what is the command you are running? Does it include --train

[Ziegan]

Yes it includes --train

the command i used is this:

python3 data2kitti.py --kaggle-dataset-path /home/miniscale/workspace/tlt-experiments/face-mask-detection-data/Kaggle\ Medical\ Mask\ Dataset/ \ --mafa-dataset-path /home/miniscale/workspace/tlt-experiments/face-mask-detection-data/MAFA\ Dataset/ \ --fddb-dataset-path /home/miniscale/workspace/tlt-experiments/face-mask-detection-data/FDDB\ Dataset/ \ --widerface-dataset-path /home/miniscale/workspace/tlt-experiments/face-mask-detection-data/Wider\ Face\ Dataset/ \ --kitti-base-path /home/miniscale/workspace/tlt-experiments/data/training/label_2/ \ --train

[ak-nv]

Closing as no activity for 15 days. Please re-open if you see the same issue.