Api results on Object detection

I am working on dot net project for detecting seven segments digits on meters using roboflow Api which takes frame or image through webcam and give Api results.

Trained my dataset through coco json format.

After deploying my project. I used the roboflow Api in my dot net project.

There are few things that I found.

  1. When checking my project on roboflow universe, with “try with webcam”.

The speed at which objects are getting is amazing and accuracy is also up to the mark.
But when I try it in my dot net project it takes 1 to 2 second to return results. Is there another way to do runtime detection in dot net instead of Api as “try with webcam” is more faster than my roboflow Api which only takes one image and returns result after a sec or two.

Since in my project I am also using webcam to detect particular frame and hit my Api. But issue is speed , the universe “try with webcam” is way faster on getting results back.

What to do to get the same speed. Is there another way or am I missing something?

  1. Secondly in my dot net project using webcam the results some times do not give accurate results but when checking my project in roboflow universe and using “try with webcam” results are much better and accurate in comparison to my project.

I tested 10 to 30 similar images on my project vs webcam on roboflow universe. Each image i.e. “ob1.png” is showing more accuracy in roboflow universe try with webcam option but same image “ob1.png” using roboflow Api has less accuracy in my dot net project.

So how to tackle it.

Here are some images in my project through webcam.

results are 14943 but per image is 014943. it takes two to five tries to get accurate.

If I try same image with roboflow universe.

Sorry one image is allowed to upload.

In robolflow universe it detects 100% accuracy of given picture while my project Api is less accurate. Here in images you can see it is missing 0 while robolfow webcam detects each object.

I have tried multiple other images and same accuracy difference found, the results are much better on roboflow universe.

Project: ocrscale
Browser: Google Chrome
This is my project link