Try CVAT with GPU pre-annotation Demo here:

Video and Image Annotation Workflow

You can annotate both video and images in Onepanel using a customized GPU optimized version of CVAT v.2.0 - an open source annotation tool.   You can use the default TF_Annotations models or upload your own to pre-load bounding boxes directly within CVAT.  

Customizing the Tensorflow model in CVAT

You can now change the model used when Run TF Annotation button is pressed by downloading your custom models into the /onepanel/input/models directory.

You will need to ensure that you add the class file ( *.csv) and the tensorflow frozen model file (*.pb).  The class file contains the classes associated with the frozen model. 

Single Project / User Annotation

If only one user will be annotating data we recommend creating a workspace within your project using the Annotation tool environment:

Outsourced Annotation Workflow

If you are using remote teams to perform annotation we recommend using using separate Projects in Onepanel to provide segmentation between users.  The process looks like this:

1.) Create separate Projects ( creating projects ) and assign the appropriate user by role:  

  • Data Annotator 1, Data Annotator 2, Data Annotator 3 etc...
  • QC 1, QC 2, etc...

2.) Create a dataset ( create a dataset ) of raw data unlabeled data -split for each labeler (i.e., datalaber 1, 2, 3, etc...)

3.) Create workspaces ( create a workspace ) using 2CPU or GPU (depending on tools being used) for each project owner.

4.) Download each dataset into the respective workspace for each Data Labeler using the SSH terminal ( see section on downloading dataset using CLI )

5.) Each Labeler can now open up CVAT using the button on the Workspace card. 

CVAT Annotation Tool:

You can also find the complete instructions for the open source version of CVAT here:  CVAT Github docs



Try the CVAT GPU Optimized Demo here:

Did this answer your question?