search

matterport / Mask_RCNN

Mask R-CNN for object detection and instance segmentation on Keras and TensorFlow
Other
24.67k stars 11.71k forks source link

detect small objects #2707

Open tbnxyzfz1 opened 3 years ago

tbnxyzfz1 commented 3 years ago

I want to detect two classes of objects in my dataset. The size of the images is 320x256. Class 1 objects are of normal size. The size of most class 2 objects is 2x2 pixels. I wonder which parameter can I change to improve the identification result of class 2 objects? I changed RPN_ANCHOR_SCALES to (2, 4, 8, 64, 128). The loss of the trained model is less than 0.7. But the trained model still can't detect class 2 objects in the test dataset. Thank you in advance.

nataliameira commented 3 years ago

@tbnxyzfz1 Do you have a lot of small objects in the scene?

tbnxyzfz1 commented 2 years ago

@tbnxyzfz1 Do you have a lot of small objects in the scene?

No, only <10 small objects in one image. Sometimes no small objects in the images.