-
Notifications
You must be signed in to change notification settings - Fork 464
Usage
- Creating Users
- Creating Categories
- Creating Datasets
- Importing Images
- Annotating Images
- Exporting Data
Admin - User of highest permission
Owner - Creator of a specific dataset
User - The account a owner or admin has shared a dataset with
Permissions | User | Owner | Admin |
---|---|---|---|
View images | ✔️ | ✔️ | ✔️ |
Create/Edit/Delete Annotations | ✔️ | ✔️ | ✔️ |
Import Annotations | ✔️ | ✔️ | ✔️ |
Share/Download/Delete Dataset | ✔️ | ✔️ | |
Add/Download/Delete Images | ✔️ | ✔️ | |
Add Images | ✔️ | ✔️ | |
Modify other users datasets | ✔️ |
Navigate to the Categories
tab found on the navigation bar, you will be presented with a create button. Here you can create as many categories as you would like.
Before we are able to annotate we must create a dataset with some categories. A Dataset contains images and is a way of separating different types of training data. If you would like more information about datasets or categories, you can click the blue information icon located beside the title.
To create a dataset start by clicking the green Create
button located under Dataset
tile. A popup with ask for a name as well as categories. Give it a unique name, select any categories that apply and click Create Dataset
.
Navigate to the datasets card click the Edit
options found in the dropdown of the dataset card. Your category should be displayed in the categories window. From here you can change and select categories.
Once you click done you have added a category to the dataset and it should be displayed in the cards footer.
In this section, we will be looking at using the file watcher methods to import images into a dataset. For Windows users, the process is slightly different and a brief overview can be found here. While the annotator is running the /datasets
directory (found at the root of the project) is being monitored for file events. Whenever an image is deleted, created or moved the corresponding event is reflected in the database.
Advantages:
- Easy to import large datasets
- No file size limit
- Simple
Disadvantages:
- Requires access to the machine running annotator
- Slightly more complicated for windows users
Navigate to the datasets directory located at the root of the project. Inside this folder, you should see folders with the names of the datasets you have created. From here simply place images inside the corresponding folder to add them to the dataset. The annotator will start adding the images to the dataset, give it a few seconds before refreshing the client. If the images do not appear in the annotator use the re-scan button found in the user interface.
Sub-folders are highly recommended as they can be used as a filtering method in the client UI.
To open the image annotation tool, click on a dataset in which the image you want to annotate is within. A viewer of all images in the dataset will be presented. If you placed the images in subfolders you can use the subdirectories badges to help filter the images by clicking on them.
Once you have found the image you would like to annotate, simply click on the image to open it in the annotator. In this section, we will briefly look at creating your first annotation. More information can be found in the Tools section on how to improve this process.
The right panel is used for managing states of your tools and annotations. This panel provides the ability too:
- Navigate to next or previous image
- Create annotations
- Change Settings
- Add metadata
- Change colors
- Delete annotations
- Search for category or annotation
- Manage settings of the selected tool
By clicking the +
icon located beside each category name, it will create an annotation in the corresponding section. You can use the eye icons to toggle the visibility of the annotation or category. The annotation highlighted in green will the layer in which the tool selected will be applied too.
Additional features are available by clicking the settings button on the annotation such adding metadata, setting a name, and changing the color.
The left panel shows all tools currently available. Please read the tool section on the wiki page for a more detailed breakdown.
Please note once you have completed annotating closing the browser will not save any annotations you have created. The following actions will trigger a save event:
- Clicking a link using the navigation bar
- Clicking the save icon in the toolbar
- Going to the next or previous image using the annotation panel
Using models such as MaskRCNN and DEXTER allow the annotation process to be semi-automated. Any model is supported as long as the user can create a endpoint which accepts an image in a post request and returns COCO formatted code containing the annotation information.
Mask-RCNN already comes setup up as a guide for implementing other models. A script can be found in the /models/
directory to download the COCO weights. If you are using any model you must point the annotator to the URL. This option can be found in the annotator window under the settings modal (click the gear icon on the left toolbar).
DEXTR is a useful tool for quickly creating segmentation. Download the DEXTR weights to activate the tool. A script can be found in the /models
directory.
Navigate to the datasets card click the Download COCO
options found in the dropdown of the dataset card. This will download a JSON file in COCO format of all annotations and images.