Missing classes name while generating label_map.pbtxt

[sainisanjay]

I have created tfrecord usingcvat/utils/tfrecords/convert.py but when i check label_map.pbtxt it missed one class name. I have 7 classes [Car Truck MotorBikes Bicycle Bus Van Pedestrian] in my xml file but it generates only 6 classes in label_map.pbtxt . In my xml has only 2 MotorBikes samples which is not generated in label_map.pbtxt . Only lower number of sample class name is missing.

item {
	id: 1
	name: 'car'
}
item {
	id: 2
	name: 'van'
}
item {
	id: 3
	name: 'truck'
}
item {
	id: 4
	name: 'pedestrian'
}

item {
	id: 5
	name: 'bus'
}
item {
	id: 6
	name: 'bicycle'
}

[zhiltsov-max]

@sainisanjay, looking at the script, I can say that the labels are collected from input annotations, so, probably, there are no motorbike bboxes. I'd recommend to use task export or Datumaro for generating annotations in different formats, as these scripts are deprecated.

[sainisanjay]

Hi @zhiltsov-max, I have verified multiple times that MoterBikes bounding box is available in the xml file:

<box label="Objects" occluded="0" xbr="-143.55864" xtl="0" ybr="825.2294" ytl="363.41858">
<attribute name="Occlusion %">0%</attribute>
<attribute name="Truncation %">0%</attribute>
<attribute name="Class">MotorBikes</attribute>
</box>

can you elaborate what do you mean by, use task export or Datumaro for generating annotations in different formats

[zhiltsov-max]

By "task export" I meant the export buttons for the task in the CVAT UI.
By "Datumaro" I meant the Datumaro project here, which have dataset conversion facilities.

[zhiltsov-max]

Probably you should take a look at --min-train parameter and set it to 0.

parser.add_argument(
        '--min-train', metavar='NUM', required=False, default=10, type=int,
        help='The minimum number of images above which the label is considered (default: 10)'
    )

[sainisanjay]

@zhiltsov-max Thanks --min-train parameter was the issue, as i was having less than 10 samples for MoterBikes class that's why it did't generate label for that.