How

[bleakie]

Tensorrt has been improved. How can I test the effect？

[hunglc007]

@bleakie u can run benchmarks.py --framework trt --weights your_trt_model

[bleakie]

Great job，but i test yolov4 is slower than AlexeyAB(yolov4)
Why?

[liuxufenfeiya]

@bleakie u can run benchmarks.py --framework trt --weights your_trt_model

this code can only output a dict that include:
'tf_op_layer_concat_11': <tf.Tensor: shape=(1, 26, 26, 3, 17), dtype=float32, numpy=
'tf_op_layer_concat_12': <tf.Tensor: shape=(1, 13, 13, 3, 17), dtype=float32, numpy=
'tf_op_layer_concat_10': <tf.Tensor: shape=(1, 52, 52, 3, 17), dtype=float32, numpy=
so what should i do next?

[liuxufenfeiya]

also,the tensorrt start takes long time,maybe 6 minutes then it can run,
i really think it has seized up and stopped.

2020-05-14 14:45:50.809331: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libnvinfer_plugin.so.6
2020-05-14 14:52:16.882567: I tensorflow/compiler/tf2tensorrt/kernels/trt_engine_op.cc:736] Building a new TensorRT engine for PartitionedCall/TRTEngineOp_2 with input shapes: [[1,26,26,3,1], [1,26,26,3,12], [1,26,26,3,4]]

[bleakie]

@hunglc007 I think the speed is a question

[iFlyyyy]

@bleakie u can run benchmarks.py --framework trt --weights your_trt_model

this code can only output a dict that include:
'tf_op_layer_concat_11': <tf.Tensor: shape=(1, 26, 26, 3, 17), dtype=float32, numpy=
'tf_op_layer_concat_12': <tf.Tensor: shape=(1, 13, 13, 3, 17), dtype=float32, numpy=
'tf_op_layer_concat_10': <tf.Tensor: shape=(1, 52, 52, 3, 17), dtype=float32, numpy=
so what should i do next?

I have the same question.Does anyone know how to use them?

[awaymeet]

Also, I got same result which is 6 size list .