skyhehe123 / SA-SSD

SA-SSD: Structure Aware Single-stage 3D Object Detection from Point Cloud (CVPR 2020)
492 stars 106 forks source link

error on RTX3090 #92

Closed vehxianfish closed 3 years ago

vehxianfish commented 3 years ago

Hi,thanks for your greak job .I use 3090 to run the program.But errors always occurs: In code:

    xg, yg, wg, lg, rg = torch.split(box, 1, dim=-1)

    xg = xg.unsqueeze_(-1).expand(N, *window_size)
    yg = yg.unsqueeze_(-1).expand(N, *window_size)
    rg = rg.unsqueeze_(-1).expand(N, *window_size)

the errors :

  File "/media/m3090/xu/detection/lidar/SA-SSD/mmdet/models/single_stage_heads/ssd_rotate_head.py", line 385, in gen_sample_grid
    yg = yg.unsqueeze_(-1).expand(N, *window_size)
RuntimeError: diff_view_meta->output_nr_ == 0 INTERNAL ASSERT FAILED at "/pytorch/torch/csrc/autograd/variable.cpp":363, please report a bug to PyTorch.

and only yg,rg have this error,xg,wg,lg don not have errors.it confused me a lot. My lib version:

pytorch=1.7.1+cu11
spconv=1.0
RTX3090

anyone meet this question?Thank you!

vehxianfish commented 3 years ago

I have solved it,Thank you!

Jackman0316 commented 3 years ago

I got the same error. Do you remember what's the reason for this? Thank you very much!

Jackman0316 commented 3 years ago

Changing .unsqueeze_(-1) to .unsqueeze(-1) solved this, Only happens in yg and rg, so weird.

vehxianfish commented 3 years ago

Changing .unsqueeze_(-1) to .unsqueeze(-1) solved this, Only happens in yg and rg, so weird.

Sorry,so later to reply you.I only remember the reastion.because pytorch version has changed a lot.you should use alternative operations.