Gorilla-Lab-SCUT/frustum-convnet

reported sunrgbd result

timetomakesense opened this issue · 4 comments

timetomakesense commented

Hi,

Thanks for your work. I want to evaluate the network on the sunrgbd dataset. I was wondering how do you evaluate the model on the SUN RGBD dataset. It seems that f pointnet does not provide 2d detection results for sun rgbd dataset. Did you train your own rgb detector on the sunrgbd dataset, or use the gt boxes?

Best

zhixinwang commented

We train faster-rcnn with the backbone network fo ResNet50 to provide the 2D detection results of SUNRGBD. We get a similar 2D detection result as f-poinnets. The reported results are based on this 2D detector.
If we use the ground truth 2D boxes, we can get a very high 3D detection mAP.

  • 👍1
timetomakesense commented

Hi zhixin,
thank you for your reply. Would you mind share your mAP on SUN RGB-D under IoU 0.5? I would like to make a comparison while cannot find that in your paper

Thanks

zhixinwang commented

I have replyed your email.

zx970505 commented

hi, i want to know how to visualize the output of SUNRGBD ? please help me @zhixinwang @timetomakesense