Turn on combined_nms: true in the pipeline config file of model, but it cause some error.

[jichenjichen]

cmd: python export_inference_graph.py --input_type=image_tensor --pipeline_config_path --trained_checkpoint_prefix --output
turn on combined_nms: true in pipeline config
use object_detection(fasterrcnn, ssd, mask_rcnn) in model_zoo checkpoint, it always cause error:
change_coordinate_frame (normlizing coordinates relative to clip_window) is not supported by combined_nms.
it maybe some cases not support by combined_nms, how can i modify pipeline config, only turn on combined_nms: true?

[jichenjichen]

I find the coordinates of boxes should be normalized, or can not use combinednms(the coordinates should be 0 to 1 when clip boxes), why not normalize in combinednms op?