search

dotnet / machinelearning

ML.NET is an open source and cross-platform machine learning framework for .NET.
https://dot.net/ml
MIT License
8.99k stars 1.88k forks source link

OBJECT DETECTION FOR SMALL OBJECT #5844

Open masgh021 opened 3 years ago

masgh021 commented 3 years ago

OBJECT DETECTION FOR SMALL OBJECT RETURN WRONG VALUE . WHY ?

michaelgsharp commented 3 years ago

Can you share more info? Either a code sample or a sample project?

masgh021 commented 3 years ago

my image is license plate and i want to recognize digits from it . but it sends wrong value , digit size are 25px * 30px . but result is not correct .

LittleLittleCloud commented 3 years ago

Did you resize the bounding boxes to fit your image size? See the following issue for more details. https://github.com/dotnet/machinelearning/issues/5772

masgh021 commented 3 years ago

@LittleLittleCloud @michaelgsharp yes i did , but it does not recognize any tiny(small) object like digits in plate or characters in text .

LittleLittleCloud commented 3 years ago

Hi @masgh021 Sounds like your task should be more of an OCR rather than object detection? The difference is

Can you share a sample picture how large the digit number is? Object detection might work if you crop around the plate and label each digit as an object? But OCR will perform better.

masgh021 commented 3 years ago

@LittleLittleCloud like this link , image i found plate , but i can't find digit number with object detection .

LittleLittleCloud commented 3 years ago

Yeah this is something that object detection might not work well with. I would try OCR instead.

masgh021 commented 3 years ago

@LittleLittleCloud which OCR do you use for c# ?(opensource) i want use for custom digit and character , it has good accuracy . i can train it easily .

briacht commented 3 years ago

@masgh021 I know a lot of people have success with OpenCV. This article may help you: https://www.tech-quantum.com/ocr-text-using-opencv-and-tesseract/

Also here's an example of a customer who uses ML.NET to detect license plates, and then they use separate technology for the OCR part: https://dotnet.microsoft.com/apps/machinelearning-ai/ml-dotnet/customers/scancam