Failed to manually run bnpipeline on x86_64

[bafu]

Steps to Reproduce

1. Run bnpipeline manually

$ python3 bnpipeline.py --pipeline-config /etc/dyda/pipelines/configs/object_detection_and_tracking_aikea.config --disable-warmup -vvv --debug

2. Run camera with input streaming from mobile

$ python3 camera-ces-with-roi-v2.py --stream-src rtsp://192.168.1.1:8080/video/h264 --fps 2 --debug

3. See exception in bnpipeline output

I added exception handling in bnpipeline.py, and got the exception

[W 200719 12:23:52 bnpipeline:74] 'Interpreter' object has no attribute 'set_num_threads'

Environment

• tensorflow-cpu 2.2.0

[bafu]

Paul also met this issue when implementing the BerryNet YOLOv4 engine.

We suspect that this is caused by TensorFlow API change (1.x -> 2.x) and are checking it.

---

The set_num_threads method disappears from tf.lite.Interpreter.

https://www.tensorflow.org/api_docs/python/tf/lite/Interpreter

---

Confirmed that it's a TensorFlow issue.

PINTO0309/Tensorflow-bin#22

The Python API of SetNumThreads is not available in current TF. The community's patch has been accepted, and this issue should be fixed in the future.