supported models for object detection

running https://github.com/NVIDIA-AI-IOT/tf_trt_models/blob/master/examples/detection/detection.ipynb in nvidia/tensorflow:19.01-p3 container with Tesla V100 16GB:

the provided detection.py allows to download and use the following models:

MODELS = {
    'ssd_mobilenet_v1_coco': DetectionModel(
        'ssd_mobilenet_v1_coco',
        'http://download.tensorflow.org/models/object_detection/ssd_mobilenet_v1_coco_2018_01_28.tar.gz',
        'ssd_mobilenet_v1_coco_2018_01_28',
    ),
    'ssd_mobilenet_v2_coco': DetectionModel(
        'ssd_mobilenet_v2_coco',
        'http://download.tensorflow.org/models/object_detection/ssd_mobilenet_v2_coco_2018_03_29.tar.gz',
        'ssd_mobilenet_v2_coco_2018_03_29',
    ),
    'ssd_inception_v2_coco': DetectionModel(
        'ssd_inception_v2_coco',
        'http://download.tensorflow.org/models/object_detection/ssd_inception_v2_coco_2018_01_28.tar.gz',
        'ssd_inception_v2_coco_2018_01_28',
    ),
    'ssd_resnet_50_fpn_coco': DetectionModel(
        'ssd_resnet_50_fpn_coco',
        'http://download.tensorflow.org/models/object_detection/ssd_resnet50_v1_fpn_shared_box_predictor_640x640_coco14_sync_2018_07_03.tar.gz',
        'ssd_resnet50_v1_fpn_shared_box_predictor_640x640_coco14_sync_2018_07_03',
    ),
    'faster_rcnn_resnet50_coco': DetectionModel(
        'faster_rcnn_resnet50_coco',
        'http://download.tensorflow.org/models/object_detection/faster_rcnn_resnet50_coco_2018_01_28.tar.gz',
        'faster_rcnn_resnet50_coco_2018_01_28',
    ),
    'faster_rcnn_nas': DetectionModel(
        'faster_rcnn_nas',
        'http://download.tensorflow.org/models/object_detection/faster_rcnn_nas_coco_2018_01_28.tar.gz',
        'faster_rcnn_nas_coco_2018_01_28',
    ),
    'mask_rcnn_resnet50_atrous_coco': DetectionModel(
        'mask_rcnn_resnet50_atrous_coco',
        'http://download.tensorflow.org/models/object_detection/mask_rcnn_resnet50_atrous_coco_2018_01_28.tar.gz',
        'mask_rcnn_resnet50_atrous_coco_2018_01_28',
    )
}

optimizing SSD_mobileNet runs without problems, however
trying

MODEL = 'ssd_resnet_50_fpn_coco'

and optimizing with

frozen_graph, input_names, output_names = build_detection_graph(
    config=config_path,
    checkpoint=checkpoint_path,
    batch_size=1
)

gives following error:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-5-dcab7238b8e0> in <module>
      2     config=config_path,
      3     checkpoint=checkpoint_path,
----> 4     batch_size=1
      5 )

/workspace/tf_trt_models/tf_trt_models/detection.py in build_detection_graph(config, checkpoint, batch_size, score_threshold, force_nms_cpu, replace_relu6, remove_assert, input_shape, output_dir)
    148                 checkpoint_path,
    149                 output_dir,
--> 150                 input_shape=[batch_size, None, None, 3]
    151             )
    152 

~/.local/lib/python3.5/site-packages/object_detection-0.1-py3.5.egg/object_detection/exporter.py in export_inference_graph(input_type, pipeline_config, trained_checkpoint_prefix, output_directory, input_shape, output_collection_name, additional_output_tensor_names, write_inference_graph)
    472       output_collection_name,
    473       graph_hook_fn=None,
--> 474       write_inference_graph=write_inference_graph)
    475   pipeline_config.eval_config.use_moving_averages = False
    476   config_util.save_pipeline_config(pipeline_config, output_directory)

~/.local/lib/python3.5/site-packages/object_detection-0.1-py3.5.egg/object_detection/exporter.py in _export_inference_graph(input_type, detection_model, use_moving_averages, trained_checkpoint_prefix, output_directory, additional_output_tensor_names, input_shape, output_collection_name, graph_hook_fn, write_inference_graph)
    383       input_shape=input_shape,
    384       output_collection_name=output_collection_name,
--> 385       graph_hook_fn=graph_hook_fn)
    386 
    387   saver_kwargs = {}

~/.local/lib/python3.5/site-packages/object_detection-0.1-py3.5.egg/object_detection/exporter.py in _build_detection_graph(input_type, detection_model, input_shape, output_collection_name, graph_hook_fn)
    351       input_tensors=input_tensors,
    352       detection_model=detection_model,
--> 353       output_collection_name=output_collection_name)
    354 
    355   # Add global step to the graph.

~/.local/lib/python3.5/site-packages/object_detection-0.1-py3.5.egg/object_detection/exporter.py in _get_outputs_from_inputs(input_tensors, detection_model, output_collection_name)
    328   preprocessed_inputs, true_image_shapes = detection_model.preprocess(inputs)
    329   output_tensors = detection_model.predict(
--> 330       preprocessed_inputs, true_image_shapes)
    331   postprocessed_tensors = detection_model.postprocess(
    332       output_tensors, true_image_shapes)

~/.local/lib/python3.5/site-packages/object_detection-0.1-py3.5.egg/object_detection/meta_architectures/ssd_meta_arch.py in predict(self, preprocessed_inputs, true_image_shapes)
    391                              [preprocessed_inputs]):
    392         feature_maps = self._feature_extractor.extract_features(
--> 393             preprocessed_inputs)
    394       feature_map_spatial_dims = self._get_feature_map_spatial_dims(
    395           feature_maps)

~/.local/lib/python3.5/site-packages/object_detection-0.1-py3.5.egg/object_detection/models/ssd_resnet_v1_fpn_feature_extractor.py in extract_features(self, preprocessed_inputs)
    155               [(key, image_features[key])
    156                for key in ['block2', 'block3', 'block4']],
--> 157               depth=256)
    158           last_feature_map = fpn_features['top_down_block4']
    159           coarse_features = {}

~/.local/lib/python3.5/site-packages/object_detection-0.1-py3.5.egg/object_detection/models/feature_map_generators.py in fpn_top_down_feature_maps(image_features, depth, scope)
    223         output_feature_map_keys.append('top_down_%s' % image_features[level][0])
    224       return collections.OrderedDict(
--> 225           reversed(zip(output_feature_map_keys, output_feature_maps_list)))
    226 
    227 

TypeError: argument to reversed() must be a sequence

Other models than the default SSD_mobileNet all throw errors during optimization.
Is there something else that needs to be cosidered when optimizing these models ?
I want to be able to optimze the tensorflow object_detection models from the tensorflow model zoo (https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md) to run performance comparisons.

Hello, we are triaging and will keep you updated.

Hello,

Per engineering, these models are fixed in TF 1.13. Could you try with 19.04 (or 19.05) version of NVIDIA containers.

regards,
NVIDIA Enterprise Support