1) Choose a Name
A descriptive name is helpful. You may change this later if needed.
2) Choose Schema (Labels & Attributes)
Choose which schema to use for these tasks. All labels available in the project are selected by default.
3) Who can Annotate?
Decide who is able to work on these tasks.
4) Who reviews? (Optional)
5) Add Fresh Data? (Optional)
In Diffgram there are three ways to load data:
- Data can be loaded during the task creation step. In this case you can click upload files.
- Data can be loaded before the task creation step. In this case, skip step 5.
- Data can be loaded after the task creation step. In this case, load data after creation to the datasets selected during step 6.
From an operational view, in step 3 you can see that a person can work on Diffgram, and setup a pipeline, simply by creating the template.
6) Select Datasets
- Select a single dataset
- Select multiple datasets
Optional: Choose what happens after Tasks are completed
Choosing copy here will copy the files to a different dataset upon completion. That dataset can in turn be watched by another Task Template.
7) UI Customization (Optional)
- Select an already customized UI
- Launch the UI editor
8) Guides/Instructions (Optional)
Guides are a way to keep important instructions top of mind while the annotation work is being done. You can define guides in your project and reuse them.
Guides can be especially relevant for:
- Technical requirements, such as tight vs looser bounding boxes
- Clarity on edge cases
- Guides are required in case of 3rd party distribution (eg Scale)
9) Advanced Settings (Optional)
Default Automation Userscript
You can choose which script shows up to annotators by default.
Relevant to multiple versions of files. If this is your first batch you can skip this section.
Use Existing (Default)
At a high level this is saying "Keep improving the same file".
New "tracking" for task information is created while work is saved directly to the original file. Using existing files automatically includes existing instances.
When to use this:
- First pass of files
- Iterative improvement of files, for example changing or adding labels
The other option is to "Isolate" in which new versions of the files will be created. It's a fresh version of the data.
If Isolation is choose, the default is to not copy existing instances.
When to use this:
- Multiple annotators on a single file
- Historical record, stronger audit trail
Use Existing Automatically Resets "Complete" Status
When a new job is launched, files set to complete will be automatically "reset".
10) Credentials/Awards (Optional)
As part of training your new annotators you may wish to create Exams. Exams award Credentials. These Credentials can be required on new tasks. This gives you an automatic way to
- Ensure basic training is complete
- Ensure data type specific training is complete
- Other considerations custom to your business
Launch the tasks!
Updated about 1 year ago