Open minju-Kang opened 11 months ago
Hi, @minju-Kang The pretrained results reported in the paper are the averages of three different splits. Thus, the pretrained results you obtained are indeed correct, and this difference should not have a significant impact.
Hi, I want to reproduce your work, but I got different pretrain performance with voxel-rcnn. I got 72.44, 27.42, 28.39 for Car, Ped, Cyc. Could you provide Voxel-RCNN checkpoint file pretrained with 1% labeled data? Thank you.