TensorRT uff SSD network h=288 w=512

AI & Data Science Deep Learning (Training & Inference) TensorRT
1

Description

We are trying to run Inference on a custom trained MobileNetV2 SSD, but when we invoke

SampleUniquePtr<IHostMemory> plan{builder->buildSerializedNetwork(*network, *config)};

on our model.uff we face this error:

Assertion numPriors * numLocClasses * 4 == inputDims[param.inputOrder[0]].d[0]’ failed.`

Our model is trained with TF 1.15 and these are the parameters:
width: 512
height: 288
channels: 3
alpha: 1.4

After the train we exported the frozen graph:

python object_detection/export_inference_graph.py \
--input_type=image_tensor \
--pipeline_config_path=pipeline.config \
--output_directory=out_dir/ \
--trained_checkpoint_prefix=model.ckpt-4000

and then we obtained .uff model using build_engine.py with these parameters:

Input = gs.create_plugin_node(
        name='Input',
        op='Placeholder',
        shape=(1,) + (3, 288, 512)
    )

    PriorBox = gs.create_plugin_node(
        name='MultipleGridAnchorGenerator',
        op='GridAnchor_TRT',
        minSize=0.2,
        maxSize=0.9,
        aspectRatios=[1.0, 2.0, 0.5, 3.0, 0.33],
        variance=[0.1, 0.1, 0.2, 0.2],
        featureMapShapes=[32, 16, 8, 4, 2, 1],
        numLayers=6
    )

    NMS = gs.create_plugin_node(
        name='NMS',
        op='NMS_TRT',
        shareLocation=1,
        varianceEncodedInTarget=0,
        backgroundLabelId=0,
        confidenceThreshold=0.3,
        nmsThreshold=0.6,
        topK=100,
        keepTopK=100,
        numClasses=8
        inputOrder=inputOrder,
        confSigmoid=1,
        isNormalized=1
    )

NOTE: If we train a squared input network (512x512x3) there are no issue performing inference

2

Please check this,
it should help