Training custom vision network using pipeline

Question

question

HankChengHanWHQWistron-5648 asked Mar 15, '22 HankChengHanWHQWistron-5648 commented Mar 24, '22

Training custom vision network using pipeline

Hi

I recently survey cloud training.
I have local yolov5 train python files and custom CNN model.
Could I upload these python file and coding pipeline, then go training and output custom model ?
I want a specific anwser. It seems azure custom vision means only use azure model but different uploading images.
Is my understanding correct or not?

Thanks a lot!

azure-custom-vision

· 2

5 |1600 characters needed characters left characters exceeded

Attachments: Up to 10 attachments (including images) can be used with a maximum of 3.0 MiB each and 30.0 MiB total.

ramr-msft · Mar 16 at 10:32 AM

@HankChengHanWHQWistron-5648 Thanks for the question. Did you export your model from custom Vision.

0 ·

HankChengHanWHQWistron-5648 ramr-msft · Mar 17 at 12:33 AM

Could you help to reply latest question?

0 ·

Answer 1 · 2022-03-16T10:48:49.31Z

ramr-msft answered Mar 16, '22 HankChengHanWHQWistron-5648 commented Mar 21, '22

@HankChengHanWHQWistron-5648 Currently you cannot upload python file to custom vision to start training and output custom model. Yes custom vision uses the azure model to train and export the model to customize and run locally.

· 1

5 |1600 characters needed characters left characters exceeded

Attachments: Up to 10 attachments (including images) can be used with a maximum of 3.0 MiB each and 30.0 MiB total.

HankChengHanWHQWistron-5648 · Mar 21 at 01:16 AM

Hello

Any update about my latest question?
I use Azure machine learning service.
Upload custom dataset and train a yolov5 s model.
I and downloaded successfully.
But now how can I run it locally?
What's the local inference example code?

Appreciate!

0 ·

Answer 2 · 2022-03-16T14:50:05.757Z

HankChengHanWHQWistron-5648 answered Mar 16, '22 HankChengHanWHQWistron-5648 commented Mar 24, '22

Hi

Thanks for reply.

So, the conclusion is that I can not write our own CNN model like mnist in train file for azure cloud training.
Only choose fix model that azure ml supply.

My next question is that I already used azure custom vision for azure yolov5 training.
I trained successfully and download a yolov5 model from azure.

I also download yolov5 source code from github.
But now I ran yolov5 inference with azure yolov5 model show error message.

 ckpt = (ckpt.get('ema') or ckpt['model']).float()  # FP32 model

KeyError: 'model'

How do I run and deploy inference locally with azure yolov5 model ?
Please guide.

Thanks

· 2

5 |1600 characters needed characters left characters exceeded

Attachments: Up to 10 attachments (including images) can be used with a maximum of 3.0 MiB each and 30.0 MiB total.

ramr-msft · Mar 24 at 07:26 AM

@HankChengHanWHQWistron-5648 Yes you can only choose the fixed model. Here is the doc to load exported model and test.

https://docs.microsoft.com/en-us/azure/cognitive-services/custom-vision-service/export-model-python#load-your-model-and-tags

0 ·

HankChengHanWHQWistron-5648 · Mar 24 at 08:22 AM

Sorry for bothering again.
It seems not have yolov5 pytorch sample code.
Is there any document relate to ?
Thanks

0 ·

question