Annotation Upload Documentation
NeuralMarker allows the user to upload annotations along with the dataset. Follow the steps provided below to do so.
- Login to the NeuralMarker tool.
- Click on the Add button. The dataset creation pop-up will appear.
- Fill in all the required details.
- Go to Add Annotations and click on it. All the supported annotation formats belonging to the selected Category Type will be listed in the drop-down menu.
- Select the annotation format you want from the drop-down menu.
- Click on the Choose File button and select the annotation files.
- To add multiple annotation files, click on the Add another annotation file icon and repeat the process.
- To remove the selected annotation file, click on the delete icon.
- Click on the Submit button and wait for your dataset to completely upload.
- After the dataset is completely uploaded, go to the Preview page or Annotation page to see the uploaded annotations.
Annotation Formats
Rectangle
1. Tensorflow COCO / COCO:
Tensorflow COCO is a JSON file containing the annotation information in key-value pairs. It has mainly five keys i.e info, licenses, images, annotations and categories.
Annotation format:
2.Tensorflow VOC / VOC:
Tensorflow VOC is an XML file. Each image has its corresponding annotation XML file. To upload VOC to NeuralMarker, all the XML files should be kept in a ZIP file and the ZIP file should be uploaded.
Annotation format:
folder - name of the folder where images are stored
filename - name of the image file
path - path where the image file is stored
width - width of the image
height - height of the image
segmented - 0 if bbox, 1 if segmentation
name - name the annotated category
xmin - left x point
xmax - right x point
ymin - top y point
ymax - bottom y point
3. Tensorflow API:
Tensorflow API is a CSV file. The first row contains the heading of the annotations that is in the order filename, width, height, class, xmin, ymin, xmax, ymax.
Annotation format:
4. Create ML:
Create ML is a JSON file containing a list of objects. Each image is an object and its annotations are a list of objects inside it.
Annotation format:
Segmentation
1. Tensorflow COCO / COCO:
The format is the same as that of Rectangle. Here inside the “annotation”, “bbox” will be empty and “segmentation” contains the segmentation values.
2. Mask RCNN:
Mask RCNN annotation file is a JSON file containing the annotation information in key-value pairs. The keys are “_via_settings”, “_via_img_metadata”, “_via_attributes”.
Annotation format:
filename - name of the image
size - size of the image in bytes
region - list of annotation objects.
shape_attributes - contains the list of x and y coordinates
all_points_x - contains the list of x coordinates
all_points_y - contains the list of y coordinates
region_attributes - contains the category information
Name (region_attributes) - category name
OCR
1. EAST
EAST annotation file is a TEXT(.txt) file. Each image has its corresponding annotation TEXT(.txt) file. To upload the EAST annotation file to NeuralMarker, all the TEXT(.txt) files should be kept in a ZIP file and the ZIP file should be uploaded. The annotation is in the order of x1, y1, x2, y2, x3, y3, x4, y4, prediction. Where “prediction” is the OCR prediction and x, y are the coordinates.
Polygon
1. Tensorflow COCO / COCO:
The format is the same as that of Rectangle and Segmentation. Here inside the “annotation”, “bbox” will be empty and “segmentation” contains the polygon values.