Wednesday, 30 June 2021

Can Mask R-CNN be used to detect specific objects in an EER diagram?

Currently I am training a model to detect binary relationships in an EER diagram using Mask R-CNN, but I am not sure the method which I am following is correct or not.

So what I basically need to do is I need to detect the binary relationships in a given EER diagram like below.

Please note news --> read --> User should also be detected.

Expected output

Expected output

So to achieve this what currently I am doing is by using VGG image annotator I am annotating a data set like below. I have trained a model with about 60 annotated images like below. I know 60 is definitely not enough. But to see is it working to even some extend I have then tried my trained model with some sample images. But the output which I got is not acceptable, because even though the model detects some of the relationships, sometimes it detects a bunch of entities at once as a binary relationship. Please refer to the output screenshot. So what I need to know is, is my approach correct or wrong. Can I get my expected results by annotating maybe around 2000 images like the image which I annotated below?

Sample annotated image

Sample annotated image

Actual output DETECTION_MIN_CONFIDENCE = 0.5

enter image description here



from Can Mask R-CNN be used to detect specific objects in an EER diagram?

No comments:

Post a Comment