Installation
The ANS Labelling Tool is built inside ANSTS and ODHUB, but you can download a standalone version of it (without licensing required) from our Web Portal.
Introduction
A sufficient training dataset for object detection includes:
- Image files
- Label files: contain all object information of each image
- Map file: contains object class IDs so the software can understand
Each object detection framework has its format for label files and map files. (Eg: YOLO has ".txt" label files and "data.yaml" map file; Pascal VOC has ".xml" label files and "label_map.pbtxt" map file)
For the ANS Labelling Tool, the label format will be ".txt" label files and "ANS_Class.json" map file
ANS Labeling Tool allows you to:
- Add, modify, and delete labels of your dataset in this format.
- Import label files from other formats (YOLO, Pascal VOC, JSON) and create a copy of them in ANS format so you can reuse your original label information.
Launch
Launching the tool will prompt you to select your Engine Type. For this context, please choose Object Detection and press OK.
Import Dataset
Browse to your dataset folder and press Current Folder to load the dataset.
The Import Data dialogue will be shown.
If your dataset does not have label files, select No and press OK to begin your labeling process.
If your dataset has label files, select Yes and choose the label format you have to import.
Press Check Label and Map File Information
If the system can locate the correct label files and map files, all label information will be displayed in the dialog box. Press OK to begin generating files in ANS format. Once your dataset is compatible with ANS, you can skip the following steps or make any necessary adjustments.
If the map file is missing, the system cannot identify the object class and will ask to generate the ANS_Class.json format. Once the map file is generated, the object class name will be labeled as "unlabeled class." Press OK to start generating files in ANS format, and then you can begin to adjust your object class.
Label Image
To label an object from an image:
- Drag and drop the cursor around the object of interest
- Enter a new object class name or select the object from the drop-down list (if any)
ANS Labelling Tool provides you with the following functions to help with your labeling process
Options
The Options menu allows you to:
- Draw mode: enable to change the cursor into drag and drop state; disable to change the cursor into bounding box selection state
- Scale to fit: Auto-rescale image to fit screen size
- Show labels: show the label text next to its bounding boxes
- Draw box color: change the color of the draw box
- Scale: Adjust image size
- Label text size: adjust label text size
- Corner dot size: adjust the bounding box's line width
Hot Keys
Press the Hot Keys Help will show all Hot Keys list in the tool
Auto-Labelling
The Auto-Labelling function (built-in version only) allows you to automatically label all images by doing AI inferencing with a pre-trained model. To start the auto-labeling process:
- Select the model to use
- Set minimum score for detection
- Press Prelabel All Files
Auto Object Detection
The auto object detection is a powerful function that can automatically highlight objects in the image by drawing bounding boxes around them, saving you time and effort on manual bounding box creation. Now, all you need to do is select a bounding box or hold Ctrl to select multiple bounding boxes, and then assign them an object class name.
Sync Labels
This function will copy all labels from the current image and paste them into the rest of the training dataset. Noted that this will overwrite all previous labels of the training dataset as well.
Augmented Images
Object Classes and Image Labels
This function allows you to manage all object classes from the training dataset.
- Bounding box color: You can change the bounding box's color for each class by clicking on the color box
- Rename object class: You can rename the object class by typing directly on its name box
- Delete object class: You can delete any object class from the dataset by pressing right-click on its name and selecting delete class.
Current Image Label Info
This function allows you to manage the bounding boxes in the current image.
- Show/hide labels of the current image by using the check box
- Change the bounding box name by double-clicking on it
- Right-click on each bounding box name to delete it from the current image
File Navigation
The file navigation option allows you to navigate through each image and do the following actions:
- You can select multiple images by pressing Ctrl or Shift button
- Delete the selected images by clicking the
icon or pressing the Delete button on your keyboard
- Search for file name with the file name filter
- Filter images by object class by clicking the drop-down list