search

matterport / Mask_RCNN

Mask R-CNN for object detection and instance segmentation on Keras and TensorFlow
Other
24.57k stars 11.69k forks source link

Data augmentation #2379

Open BS-98 opened 3 years ago

BS-98 commented 3 years ago

Hi,

I have read many issues about data augmentation and I am a little bit confused. Could tell me if I can add all augmentation techniques from imgaug library or I cannot use some of these like flip, rotate (geometric) beacuse of masks?

Maybe you can advise if use Sequential, OneOf or something different method and particular techniques like as blur, contrast, add, multiply?

I would be grateful for any help.

Shakesbeer333 commented 3 years ago

Hi @BS-98,

yes geometric data augmentation can be tricky. If you apply such an augmenter, you need to make sure that the the masks are transformed in the corresponding way. You can apply augmenters such as contrast, add, multiply without any worries as they don't effect the masks.

For a solution see: #2369

ashwin-999 commented 3 years ago

@BS-98 Maybe, this helps clarify? https://github.com/matterport/Mask_RCNN/issues/2018#issuecomment-787080780

This comment is only half-true https://github.com/matterport/Mask_RCNN/issues/281#issuecomment-377485312

denashamss commented 3 years ago

Hi everyone, I have applied data augmentation but I dont know how can I print Number of dataset images after training the model using augmentation?