Model missing in Computer vision

Hi Punchi

Sorry for the trouble on email notification.

Custom Vision from computer vision was depreciated in favor of Custom vision portal

Features like Azure AI Image Analysis 4.0 Custom Image Classification, Custom Object Detection, and Product Recognition preview API from Azure compute vision has been retired on March 31 2025

So, API calls to model with these services are failing.

Reference -

https://free.blessedness.top/en-us/azure/ai-services/computer-vision/how-to/model-customization?tabs=studio#create-a-new-custom-model

https://free.blessedness.top/en-us/answers/questions/2259020/training-custom-vision-model-fails-with-unknown-er

Recommendation

You can create a new model in custom vision portal to re-create the same model with same storage dataset

Minimum 5 images is required for image classification (10 for object detection)

Reference -

Getting started with custom vision portal

Input file requirement

Please let us know if it helps you migrate to custom vision

Thank you for understanding.

Hi Punchi

AutoML need labelled images

https://free.blessedness.top/en-us/azure/machine-learning/how-to-prepare-datasets-for-automl-images?view=azureml-api-2&tabs=python

If they want to avoid re-tagging the images, you can segregate the images in respective folder names as per their tags (2000 tags) and use Yolo notebook for training the model. You can deploy that model as online or batch endpoint on consumption.

References

Image Classification - Ultralytics YOLO Docs

IMAGE CLASSIFICATION with Yolov8 custom dataset | Computer vision tutorial

Deploy Machine Learning Models to Online Endpoints - Azure Machine Learning | Microsoft Learn (customer need to do prediction script score.py to deploy to online endpoint)

What are batch endpoints? - Azure Machine Learning | Microsoft Learn

Author scoring scripts for batch deployments - Azure Machine Learning | Microsoft Learn (Scoring is 2 step process, load the yolo v11 model and pass the images for prediction)

Sample scoring script for online/Batch endpoints

def init():    global model
   # AZUREML_MODEL_DIR is an environment variable created during deployment    # The path "model" is the name of the registered model's folder    
model_path = os.path.join(os.environ["AZUREML_MODEL_DIR"], "model")
   # load the model    model = load_model(model_path)
 
import pandas as pd from typing import List, Any, Union
def run(mini_batch: List[str]) -> Union[List[Any], pd.DataFrame]:    results = []
   for file in mini_batch:        (...)
   return pd.DataFrame(results)

Hope it saves the re-tagging effort for all 2000 tags images.

Thank you