Custom Vision Classification Model Failure about vision-ai-developer-kit HOT 9 OPEN

@santle - Please confirm if you have followed the steps correctly from this guide - https://azure.github.io/Vision-AI-DevKit-Pages/docs/Deploy_Model_IoT_Hub/
Also try to stream the video in RTSP, refer this guide - https://azure.github.io/Vision-AI-DevKit-Pages/docs/RTSP_stream/

from vision-ai-developer-kit.

Right. We followed the above steps and pushed the latest containers to the camera:

/ # docker ps
CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES
f575ec24919b mcr.microsoft.com/aivision/visionsamplemodule:webstream_0.0.13-arm32v7 "npm start" 17 hours ago Up 17 hours WebStreamModule
4eec2b17aaa7 mcr.microsoft.com/azureiotedge-hub:1.0 "/bin/sh -c 'echo "$…" 17 hours ago Up 17 hours 0.0.0.0:443->443/tcp, 0.0.0.0:5671->5671/tcp, 0.0.0.0:8883->8883/tcp edgeHub
592f4db7caaa mcr.microsoft.com/aivision/visionsamplemodule:latest "python3 -u ./python…" 17 hours ago Up 17 hours AIVisionDevKitGetStartedModule
c237ee314940 mcr.microsoft.com/azureiotedge-agent:1.0 "/bin/sh -c 'exec /a…" 17 hours ago Up 17 hours edgeAgent

After this, we added customvision.ai trained classification model (sample.VAIDK.zip -a publicly accessible blob file) to the Module Identity Twin of AIVisionDevKitGetStartedModule:

{
"deviceId": "AICamera",
"moduleId": "AIVisionDevKitGetStartedModule",
"etag": "AAAAAAAAAAU=",
"deviceEtag": "Mjg5MjQwMDYy",
"status": "enabled",
"statusUpdateTime": "0001-01-01T00:00:00Z",
"connectionState": "Connected",
"lastActivityTime": "2021-02-10T19:28:14.8984726Z",
"cloudToDeviceMessageCount": 0,
"authenticationType": "sas",
"x509Thumbprint": {
"primaryThumbprint": null,
"secondaryThumbprint": null
},
"modelId": "",
"version": 8,
"properties": {
"desired": {
"ModelZipUrl": "https://sample.VAIDK.zip",
"TimeBetweenMessagesInSeconds": 12,
"ObjectsOfInterest": "["ALL"]",
"ShowVideoOverlay": true,

Camera logs show twin updates:

Twin callback called with:
updateStatus: PARTIAL
payload: {"ModelZipUrl":"https://sample.VAIDK.zip","TimeBetweenMessagesInSeconds":12,"ObjectsOfInterest":"[\"ALL\"]","ShowVideoOverlay":true,"Bitrate":"1.5Mbps","Resolution":"1080P","VideoOverlayConfig":"inference","FrameRate":30,"HdmiDisplayActive":true,"VideoAnalyticsEnabled":true,"ShowVideoPreview":true,"Codec":"AVC/H.264","$version":5}
INFO:root:timestamp=59265701309
INFO:root:id=2
INFO:root:label=person.
INFO:root:confidence=77
INFO:root:Position(x,y,w,h)=447.936,333.936,1042.944,733.968
INFO:root:timestamp=59265734642
INFO:root:id=1
INFO:root:label=person.
INFO:root:confidence=74
INFO:root:Position(x,y,w,h)=453.888,308.988,1067.904,751.896
INFO:root:timestamp=59265801309
INFO:root:id=2
INFO:root:label=person.
INFO:root:confidence=74
INFO:root:Position(x,y,w,h)=453.888,308.988,1067.904,751.896
INFO:root:timestamp=59265834642
INFO:root:id=1
INFO:root:label=person.
INFO:root:confidence=70
INFO:root:Position(x,y,w,h)=440.832,308.988,1067.904,751.896
INFO:root:timestamp=59265934642
INFO:root:id=1
INFO:root:label=person.
INFO:root:confidence=70
INFO:root:Position(x,y,w,h)=440.832,330.912,1035.84,740.988
INFO:root:timestamp=59266001309
INFO:root:id=2
INFO:root:label=person.
INFO:root:confidence=70
INFO:root:Position(x,y,w,h)=440.832,330.912,1035.84,740.988
INFO:root:timestamp=59266034642
INFO:root:id=1

But, we are not seeing classification results on the display/ within IoThub telemetry.

from vision-ai-developer-kit.

@drwill-ms Hi David, Can you please look into this issue?

from vision-ai-developer-kit.

I don't work on the Vision AI dev kit any more, but I asked around for someone to pick it up.

from vision-ai-developer-kit.

@drwill-ms Thanks david.

from vision-ai-developer-kit.

@santle - Just following up on this ticket. Please advise do you still face this reported issue.

from vision-ai-developer-kit.

I am having the same issue on my VAIDK. I have updated my module twin using several different models trained and exported from CustomVision.ai. They were trained with the General Compact domain and exported in VAIDK format. I have loaded them to a public blob storage container and updated the module twin. After the update, monitoring the getting started module logs on the device shows that the update occurred successfully but I never get any messages sent to IoT Hub. My log output looks virtually identical to that posted by @santle.

As at test, I pulled the model files for the getting started model from this repo, zipped them and put the .zip into my public blob container. Updating my module twin to this path loads the model successfully and i get messages sent to IoT hub as expected.

from vision-ai-developer-kit.

Also having this same issue anything other than the default model will not detect a labeled object something has changed with the CustomVision.ai. models ?

Does anyone at Microsoft even support this device anymore ?

from vision-ai-developer-kit.

Also having the same issue when using a customvision trained model. Using General (compact) domain.

from vision-ai-developer-kit.