Is Mask R-CNN available in Matlab?

Faster R-CNN is the backbone of Mask R-CNN. However, adding the needed pixelwise classification to Faster R-CNN causes an issue with training using the trainFasterRCNNObjectDetector function.

Has anyone built Mask R-CNN in Matlab, and found a way to train it?

[Wassim Dhaouadi]

Please update us if you figured this out. I and many others would be very interested to know how you did it.

-Thank you

Answer 1

[Athul Prakash]

Matlab doesn't have a pre-trained Mask RCNN network as of now.

You may write your own training script for the custom network you mentioned.

However, I recommend pouring through File Exchange - you might find someone's implementation over there.

Answer 2

[Tohru Kikawada]

Mask RCNN is available from R2020b.

https://jp.mathworks.com/help/vision/ug/getting-started-with-mask-r-cnn-for-instance-segmentation.html

[L]

Dear Tohru,

I have tried to follow the demo from this link https://www.mathworks.com/help/vision/ug/example-InstanceSegmentationUsingMaskRCNNDeepLearningExample.html , but I can't train the mask rcnn on my on dataset, I get various errors. I have labeled the data and I have converted it to json file.

I would like to train the net with my own small dataset in order to learn the procedure.

Is there a demo where can we all see the training on custom dataset (small dataset with few images)?

I cannot find the example with custom dataset. Is it available?

The example from the provided link is related to COCO, prepaired dataset and it is difficult to follow. It is confusing since it uses helper functions and aready prepared images and annotation.