Image Annotation Using Makesense.ai
Capturing images for data collection is easy; what comes next is challenging. Image annotation or labeling is one of the most time-consuming and labor-intensive steps of data preprocessing in machine learning and computer vision tasks.
Image annotation means adding labels or markings to images to identify and define objects or regions of interest. It typically includes bounding boxes and labels to indicate the type of object or category.
You can annotate your images using makesense.ai, a user-friendly image annotation platform. It provides a straightforward and intuitive interface for labeling objects in images. While VIA (VGG Image Annotator) is an alternative, makesense.ai offers a simpler and more streamlined experience for image annotation tasks.
Image Annotation Instructions for Makesense.ai
- Go to the website: makesene.ai and click on Get Started on the bottom right.
2. You’ll be taken to a new window. You can click on Drop images or drag the images from your computer directly.
3. After dragging or uploading the images, click on Object Detection.
4. You’ll be asked to define your labels. You can add one or many, depending on the number of classes you’ll define. In my case, I’m just adding one label name, i.e. Anthracnose, as shown on the image below.
5. Click on the image you want to begin annotating on the left side. Then, select the shape from rectangle, point, line, and polygon. Generally, polygon shape is preferred for instance segmentation in CNN. I’ve selected polygon for my labeling as shown below:
6. Label appropriately for all objects. I prefer using the Huion Drawing Tablet for labeling. But you can use mouse as well.
7. You can import more images to the existing library by clicking on the Import Images button under the Actions tab on the top left.
8. A magnifying icon with a + and — sign is for zooming in and out. A hand icon is for dragging the images when zoomed; and an icon next to the hand icon is the cursor crosshair icon, which, when turned on, changes the appearance of the cursor to a cross-hair. With this mode on, the shape of the cursor changes into a cross-hair, which can be moved around the screen to target and select an element, such as a button, link, or image.
9. You can export the file after annotations are complete by clicking on Export Annotations under the Actions tab.
10. You can choose two formats (VGG JSON and COCO JSON) to export the file, as shown below.
Viewing JSON File in Notepad++
You can open the downloaded JSON file in Notepad++ to see the features of the annotations. In order to align the data so that it is easily readable, click on Plugins, then click on JSTool, and select JSFormat as shown in the image below.
The annotated images will be organized and displayed in the following way:
{ ‘filename’: ‘Image.jpeg’, ‘regions’: { ‘0’: { ‘region_attributes’: {}, ‘shape_attributes’: { ‘all_points_x’: […], ‘all_points_y’: […], ‘name’: ‘polygon’}}, … more regions … }, ‘size’: 100202 }
We mostly care about the x and y coordinates of each region. You can load the JSON file in Python and train your model.
Shortcut Keys for makesense.ai:
- CTRL + mouse-wheel scroll up/down: Zoom in/out.
- Mouse-wheel scroll up.down: move the image in vertical direction.
- SHIFT + mouse-wheel scroll up/down: move the image in horizontal direction.
- ESC: terminate annotation and start new.
- DEL: delete annotated object that is selected.
