The KITTI Vision Benchmark Suite

We have proposed a SLAM method which aims to
improve the performance in dynamic environment. On
the KITTI dataset, our method uses the monocular
images and point clouds to complete task. The
virtual vector, which is a vector in 3-D space
without a physical entity, is first constructed.
Then, the moving objectives can be identified with
the virtual vectors and be deleted in the whole
SLAM pipeline. After deleting those moving
objectives, our SLAM method can extract feature
points only in the static area and thus achieve a
higher accuracy.

The Mask R-CNN built by PyTorch uses the pretrained
weights.

No.