matterport / Mask_RCNN

Mask R-CNN for object detection and instance segmentation on Keras and TensorFlow
Other
24.73k stars 11.71k forks source link

Training the network with object detection dataset #1902

Open danielbispov opened 4 years ago

danielbispov commented 4 years ago

Hello!

I would like to know if it is possible to train the network on a dataset that is annotated only with bounding boxes. I want the network to be able to predict only bounding boxes. Would you please help me understand if that is possible and what would be the way to do so?

Thank you for your time,

marcfielding1 commented 4 years ago

I dunno if you can or not, what's the dataset consist of out of interest?

danielbispov commented 4 years ago

It is a camera trap images dataset representing 20 species with bounding boxes included. I want to use Mask RCNN as a relevant baseline for the object detection task. Still, I wonder if i should use an algorithm that's aimed for that task exclusively and that is state-of-the-art instead of this one.

basedrhys commented 4 years ago

The detectron2 repo has some good baseline models that work for bounding-box only datasets, you might like to check that out.

cdzhu commented 4 years ago

Hello!

I would like to know if it is possible to train the network on a dataset that is annotated only with bounding boxes. I want the network to be able to predict only bounding boxes. Would you please help me understand if that is possible and what would be the way to do so?

Thank you for your time,

of course you can!you just need to change the network ,delete the mask parts,and change the data_genetator.

sarthakforwet commented 4 years ago

@cdzhu I have a dataset with images and bounding boxes only . How do i input the bounding boxes to the model ? Where can i find data_generator?