AWS Tutorials - Create Custom Models using Amazon Rekognition Custom Labels

[@tushartyagi4087]

"Permissions have not been granted on an S3 resource." I'm getting this error while training the model. How to fix it??

[@AWSTutorialsOnline]

When you use custom labels for the first time, it creates S3 bucket and it uses it for the project purpose. Rekognition has required permission on the S3 bucket. did you successfully passed that step?

[@roidyrobertoalvarezgomez912]

models can be trained for the specific recognition of text fields in identifications?

[@AWSTutorialsOnline]

ok so you want to build a model which can recognize text in the picture. It might work. You can do a quick test. Meanwhile, Amazon Rekognition also has a text detect API which can help do it without building any model. Refer - boto3.amazonaws.com/v1/documentation/api/latest/reference/services/rekognition.html#Rekognition.Client.detect_text Try that one as well.

[@caankitrmehta2281]

So i used aws textract and checked my application form. It did not fetch all the data accurately from pdf. How to customise the extraction so next time same format comes aws textract will not make same mistake

[@stanislavhinov4225]

Hello, how can i test all of the images in a bucket? Or even multiple images?

[@AWSTutorialsOnline]

In the detect_custom_labels method, you can only pass one S3 object at a time. you can keep all your images in one bucket. You then use S3 API to list through the objects in the s3 bucket and perform detection one by one.

[@stanislavhinov4225]

Do you have a tutorial on how can we create a mobile frontend for our model in Amazon Rekognition Custom Labels?

[@AWSTutorialsOnline]

You can use Amazon Amplify and ReactJS for the frontend development.

[@chournsolidet5406]

great content. However, My app is available for clients to search products by image. should we keep starting, using, and stopping every time a client search ?

[@AWSTutorialsOnline]

If you are predicting label on the image during search then you need to keep model running for the quick response. If you start model and then search; it will be good use experience as model start takes quite sometime.

[@chournsolidet5406]

@@AWSTutorialsOnline thank you. One more thing is that I need a feature of searching product by image. I was researching for aws rekognition, so i could get predicted label and pass it to query products from my api, however i came across aws visual search (aws.amazon.com/campaigns/buildon-nordics/visualsearch/ ) which uses aws SageMaker, something like this. and I don't know which one i should go for

[@chournsolidet5406]

@@AWSTutorialsOnline and aws visual search is way beyond my understanding

[@AWSTutorialsOnline]

I am not aware of visual search. Looks interesting. But the other approach will work for sure. From Image, you generate predicted labels and those labels are used to search in your image database (assuming you image database has images and labels attached to it)

[@chournsolidet5406]

@@AWSTutorialsOnline yeah, i can get label to query my products by my api. but now when i try using the model from my local machine , i got sth like 'Cross region access not allowed'

[@caankitrmehta2281]

How to make customise lable (annotations) in a form so that next time similar form comes application can extract those particular information

[@liveyourlife998]

How to reduce the starting time of the model??? any idea please regarding it.

[@AWSTutorialsOnline]

Hi, I don't think you can control model start time. It depends on model size and complexity. It is recommended to schedule model run if you are using model in batch. When starting the model, you can configure number of inference to enhance transaction pers second throughput.

[@liveyourlife998]

@@AWSTutorialsOnline if number of inference increased can the model start bit early?

[@AWSTutorialsOnline]

number of inference can increase TPS throughput but not model start time. Think about scheduled model start strategy to optimize the time.