TensorRT 4.0 fails to convert TensorFlow object detection model

Hi, when I try to convert a tensorflow object detection model (ssd_resnet50_fpn_coco: https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md )to uff model using convert-to-uff command, it fails with error: “AttributeError: ‘int’ object has no attribute ‘as_numpy_dtype’”. Then I try to convert by uff.from_tensorflow_frozen_model, another error is reported: “NameError: name ‘UffException’ is not defined”.

Here is my conversion command:

convert-to-uff tensorflow --input-file ssd_resnet50_fpn.pb -o model.uff -I="image_tensor,,float32,640,640,3" -O="num_detections,detection_classed,detection_boxes,detection_scores"

After checking the node name list produced by performing convert-to-uff -l , I find the output node name which should be “num_detection, detection_classes, detection_boxes, detection_scores” is replaced by an odd name “Const6_1539589367_419572”. But if I remove -O options in previous command, conversion succeeds. It seems that the output nodes have been modified during uff model conversion, anyone could help explain why this happens? Thanks!