search

matterport / Mask_RCNN

Mask R-CNN for object detection and instance segmentation on Keras and TensorFlow
Other
24.52k stars 11.68k forks source link

How to create the Dataset class to train Mask RCNN with my own dataset? #1625

Open chrigui94 opened 5 years ago

chrigui94 commented 5 years ago

Hello, I have a task of image segmentation using mask rcnn and I found a problem to train the model using my own images which contains only one class, I have clinical images of the body that contains a part of a person body. I am trying to create the BodyDataset class following the CocoDataset class example. is it the right way to train mask rcnn on my own dataset. do I need annotations of my images?

genausz commented 5 years ago

You can reference example/balloon codes and use VIA annotation tools to create your dataset and train

chrigui94 commented 5 years ago

@genausz thank you, can you give an idea what is the minimum number of images I have to annotate using VIA? because I have a dataset of 1000 images to train.

parkill commented 5 years ago

I think the more, the better. I learned by using 50 datasets as a test. Then he recognized the results of his own significance. Epochs = 30 , STEPS_PER_EPOCH = 100

Using a translator, the sentence may be a little strange.