Custom VIsion Model Training isn't working

Question

Custom VIsion Model Training isn't working

Kimberly Jones 20

I created 2 different projects in hopes to train my model.

I uploaded vegetables, tagged them, have 5 images in each tag, and when I select train, I received the same error message "Your project can't be trained just yet. Make sure to have 2 tags, 5 images"

Below are my current settingsScreenshot 2025-07-31 at 12.11.42 AM.png

Manas Mohanty 8,150 Microsoft External Staff Moderator

Hi Kimberly Jones

We are still waiting to hear back from you.

Here is a python script that checks for working images and saves them in .png format.

import os
from PIL import Image, UnidentifiedImageError
# Define the source and destination directories
source_dir = 'images'  # Replace with your actual image folder path
destination_dir = 'healthy_images'
# Create the destination directory if it doesn't exist
os.makedirs(destination_dir, exist_ok=True)
# Iterate through all files in the source directory
for filename in os.listdir(source_dir):
    file_path = os.path.join(source_dir, filename)
    try:
        # Open the image file to check if it's valid
        with Image.open(file_path) as img:
            # Convert and save the image in PNG format to the destination directory
            clean_filename = os.path.splitext(filename)[0] + '.png'
            img.save(os.path.join(destination_dir, clean_filename), 'PNG')
    except (UnidentifiedImageError, OSError):
        # Skip corrupt or unreadable images
        print(f"Corrupt or unreadable image skipped: {filename}")

Please take a minute to accept anyone of the below answer if it helped.

Thank you

Manas Mohanty 8,150 Reputation points Microsoft External Staff Moderator

2025-08-06T02:24:27.61+00:00

Hi Kimberly Jones

We could not hear from you. Hope you appreciate the inputs.

Please let us know if the issue persists for you.

Thank you

Accepted answer

1 additional answer

Your answer

Manas Mohanty 8,150 Reputation points Microsoft External Staff Moderator

2025-08-05T01:57:46.21+00:00

Hi Kimberly Jones

We are still waiting to hear back from you.

Here is a python script that checks for working images and saves them in .png format.

import os from PIL import Image, UnidentifiedImageError # Define the source and destination directories source_dir = 'images' # Replace with your actual image folder path destination_dir = 'healthy_images' # Create the destination directory if it doesn't exist os.makedirs(destination_dir, exist_ok=True) # Iterate through all files in the source directory for filename in os.listdir(source_dir): file_path = os.path.join(source_dir, filename) try: # Open the image file to check if it's valid with Image.open(file_path) as img: # Convert and save the image in PNG format to the destination directory clean_filename = os.path.splitext(filename)[0] + '.png' img.save(os.path.join(destination_dir, clean_filename), 'PNG') except (UnidentifiedImageError, OSError): # Skip corrupt or unreadable images print(f"Corrupt or unreadable image skipped: {filename}")

Please take a minute to accept anyone of the below answer if it helped.

Thank you
Manas Mohanty 8,150 Reputation points Microsoft External Staff Moderator

2025-08-06T02:24:27.61+00:00

Hi Kimberly Jones

We could not hear from you. Hope you appreciate the inputs.

Please let us know if the issue persists for you.

Thank you

Answer 1

Hello Kimberly Jones,

It sounds like you’re hitting a common issue with Azure Custom Vision’s training requirements. Based on your description, even though you have two tags with five images each, you’re still seeing the error. Here are some steps to troubleshoot:

1.	Check the Tag Assignment

•	Ensure each of your images is correctly tagged. Sometimes, images may appear uploaded but not actually be associated with a tag. Double-check by selecting each tag and confirming you see five unique images under each.

2.	Tag Minimums

•	The platform requires at least two different tags, each with at least five uniquely tagged images. If an image is accidentally tagged with both labels, it might not count correctly. Each tag should ideally have its own set of five distinct images.

3.	Image Formats & Upload Issues

•	Ensure all images are in a supported format (like JPG or PNG) and aren’t corrupted. Try re-uploading a few to refresh the project.

4.	Project Type

•	Double-check your project settings—are you creating a “Classification” (Single Tag per Image or Multi-tag) or “Object Detection” project? The requirements might vary slightly.

5.	Refresh & Re-Train

•	Sometimes, there’s a caching issue. Refresh the browser, log out and back in, or even create a new project as a test to see if the error persists.

6.	Limits & Quotas

•	If your account is on a free tier, check if you’ve hit a training quota.

If you’ve checked all of the above and still face the issue, try deleting the project and recreating it step-by-step, uploading images and tagging as you go. Sometimes this resets underlying problems. If the problem continues, reach out through the Azure portal’s support options—there might be a backend issue or service glitch.

Best Regards,

Jerald Felix

Answer 2

Manas Mohanty 8,150 Microsoft External Staff Moderator

Hey Kimberly Jones

As per screenshot, you are using FO tiers of custom vision.

Free Tier -TrainProject (API calls Per Second)-2

SO tier - -TrainProject (API calls Per Second)-10

You have around 94 images which is still under quota limit of 5000 images for free tier

Please check the image extension or any corrupt images - We can leverage a python script to find any corrupt images which are not in support image format and remove the corrupt image or change the image extension Accepted image types of- jpg, png, bmp, gif Please use two tags and 10 images each in each tag for image classification and 15 images in each tag if object detection for find out the difference and cause of the issue.
You can find minimum height, image and resolution criteria here in below document.

https://learn.microsoft.com/en-us/azure/ai-services/custom-vision-service/limits-and-quotas

Could you share 5 images for two tags in private message to replicate the issue if the issue persists.

Thank you

Kimberly Jones 20 Reputation points

2025-08-06T02:44:53.07+00:00

I was able to just go to Python, write code to connect to CV project, view images per tags, then train from python code.

Thank you
Manas Mohanty 8,150 Reputation points Microsoft External Staff Moderator

2025-08-06T02:50:21.0666667+00:00

Great to hear that Kimberly Jones , you were able to train from Python SDK.

Can Mark it "Resolved original poster" then.

Curious that if the issue persists on Portal UI after checking sanity of Images. Need to inform PG if the issue on UI backend.

Share via

Custom VIsion Model Training isn't working

1 additional answer

Your answer