Coral TPU model change automatically after 120sec

[bbo76]

Area of Concern

• Server
• Behaviour of one or more Modules [ObjectDetectionCoral]
• Installer
• Runtime [e.g. Python3.7, .NET]
• Module packages [e.g. PyTorch)
• Something else

Describe the bug
I call object detection API with custom model 2 time with +120sec between and the second time, the model used wasn't YOLO but default model.

API called: http://XXX:32168/v1/vision/custom/YOLOv5

server log output:
Response rec'd from Object Detection (Coral) command 'custom' (...6e9f4a) ['Found car, person, car'] took 280ms
11:48:39:objectdetection_coral_adapter.py: WARNING:root:No work in 120.0 seconds, watchdog shutting down TPUs.
11:49:47:Response rec'd from Object Detection (Coral) command 'custom' (...8c1757) ['No objects found'] took 2751ms
11:50:02:objectdetection_coral_adapter.py: Object Detection (Coral) started.
11:50:02:objectdetection_coral_adapter.py: no TPU interpreters: Falling back to CPU detection
11:50:02:objectdetection_coral_adapter.py: No interpreter found. Recreating.
11:50:02:objectdetection_coral_adapter.py: Model change detected. Forcing model reload.
11:50:02:objectdetection_coral_adapter.py: Timeout connecting to the server

As you can see, first time the model detect person car etc... and second time (same query) nothing found and the model was changed.

Expected behavior
Keep the model configured on the start of the module.

Your System (please complete the following information):

• CodeProject.AI Server version: 2.6.2]
• OS:Docker
• CoralUSB TPU

[ChrisMaunder]

USB or PCIe Coral?

[bbo76]

USB or PCIe Coral?

USB

[Jensilein]

I experience basically the same behaviour on Windows. Whereas on Linux (Ubuntu) it works fine.

[scho0ck]

Same issue on Windows with PCIe Coral Yolov5s model and CPAI 2.8