At 308, updated object prediction information is output. The updated object prediction information is based on the adjusted object prediction information. In some embodiments, the updated object prediction information includes the adjusted object prediction information, and is output by the client application to a server such as the annotation engine. In some embodiments, the initial coordinate information and/or initial classification information of adjusted objects is replaced or modified by the adjusted object prediction information to generate the updated object prediction information. In some embodiments, a server such as the annotation engine updates the object prediction information and outputs the information. The information can be sent to storage, to the requester's device, to another module of the annotation platform such as an aggregator to be aggregated with other annotator users' annotations, to an ML model for training, to a separate application, or to any other appropriate destination.

In some cases, there is no applicable ML model available initially. In such cases, one or more human annotator users annotate a set of sample images, and use the annotation results to train an ML model. Once an ML model such as 204 becomes established, process 300 can be invoked to assist future annotation by annotator users.