function readOnly(count){ }
Starting November 20, the site will be set to read-only. On December 4, 2023,
forum discussions will move to the Trailblazer Community.
+ Start a Discussion
Nicolas GUIGUINicolas GUIGUI 

Can Einstein vision determine if an image contains 2 (or moreà) image in my dataset ?

Hi everybody !

I train with the api Einstein vision and I have a business case:
Customers send us receipt and I want to use Einstein vision to determine all lines in the receipt.

If I have a custom dataset with all possible lines that I can found: Can Einstein vision determine all product in the receipt and get an API return like this:
{
"probabilities": [
{
"label": "Product 1",
"probability": 0.97554934
},
{
"label": "Product 2",
"probability": 0.974450686
},
{
"label": "Product 3",
"probability": 0.000000686
}
],
"object": "predictresponse"
}

So, this response indicates me that I have product 1 and 2 in the receipt.

Thanks a lot for your return !

Regards
Best Answer chosen by Nicolas GUIGUI
Sachin RamachandranSachin Ramachandran
Hi Nicolas,

Yes, it is possible, you can achive this by choosing the dataset type as "image-multi-label" while creating your dataset, more on this refer - https://metamind.readme.io/docs/create-a-dataset-zip-async
The diference between dataset type - "image" and "image-multi-label" , in the former the probability across all the class labels sums to one, meaning that it assume the class label are mutually exclusive & tries to find one label per image. In the later ( "image-multi-label" ie) the class label are considered to be independent of each other. The returned probability represent if the class label is present in the image or not (& hence sum of all probability doesn't sum up to one)
Hope this clarifies, Please feel free to reach out if you have more questions.

Regards,
Sachin
Einstein.ai