Mask_rcnn: Train Mask RCNN from scratch

[ojoaobrito]

Just a quick question, I want to train this implementation of Mask RCNN from scratch (i.e. without using the COCO or Imagenet weights). How can I do that?

[brahimharounhassan]

Just a quick question, I want to train this implementation of Mask RCNN from scratch (i.e. without using the COCO or Imagenet weights). How can I do that?

I would also like, it can help us a lot as a student.

J'aimerais aussi, cela peut nous aider beaucoup en tant qu'étudiant.

[sadmam]

This would in fact be great to know.

[JotaCoves]

Are you asking how start with random weights?
If is this, if you comment the line where model.load_weights() do that.

Just a quick question, I want to train this implementation of Mask RCNN from scratch (i.e. without using the COCO or Imagenet weights). How can I do that?

Are you asking how start with random weights?
If is this, if you comment the line where model.load_weights() do that.

[satishhiremath]

I think we can train MRCNN from scratch. I tried by making train "all" layers while training. But we need a huge amount of train data to get better results as network is too deep.
As there is always a space to learn, looking forward to knowing if my understanding is correct.

[ojoaobrito]

Are you asking how start with random weights?
If is this, if you comment the line where model.load_weights() do that.

Just a quick question, I want to train this implementation of Mask RCNN from scratch (i.e. without using the COCO or Imagenet weights). How can I do that?

Are you asking how start with random weights?
If is this, if you comment the line where model.load_weights() do that.

That did the trick! Thank you very much! @JotaCoves