Thanks for following up with us. Apologies for setting your expectation to have this in August. As you know with software development, many tasks are moving targets. That being said, we did add it to our road map for the year as part of a larger task of improving support for object detection architectures, but it has not been started yet.
At the moment, the solution continues to be to choose a different model architecture.
Also, for the FasterRCNN network there is a bit more we will need to support beyond ROIAlign. There are Scatter ops and shape logic that are also causing import issues.