The problem with the test

[Ange1ika]

The problem with the test

if run
torchpack dist-run -np 8 python tools/test.py configs/nuscenes/det/transfusion/secfpn/camera+lidar/swint_v0p075/convfuser.yaml pretrained/bevfusion-det.pth --eval bbox
he falls out somewhere, into an endless cycle - it is unclear what he is doing...

I tried to do debugging, at the last it outputs 185 and loads checkpoints

[zhijian-liu]

Could you try running the evaluation on a single GPU to see if the same issue still persists? Thank you.

[Ange1ika]

I'm sorry that I took a long time to answer, now I will answer every day, because there is an urgent need to complete the project. There is such an error hanging now:

Thanks for the answer

[Ange1ika]

[Ange1ika]

torchpack dist-run -np 8 python visualize.py /configs/nuscenes/seg/fusion-bev256d2-lss.yaml --model.encoders.camera.backbone.init_cfg.checkpoint pretrained/swint-nuimages-pretrained.pth --load_from pretrained/bevfusion-seg.pth

torchpack dist-run -np 1 python tools/visualize.py configs/nuscenes/seg/lidar-centerpoint-bev128.yaml --model.encoders.camera.backbone.init_cfg.checkpoint pretrained/swint-nuimages-pretrained.pth --load_from pretrained/lidar-only-seg.pth
Can you tell me what the errors are?

torchpack dist-run -np 2 python tools/visualize.py configs/nuscenes/seg/fusion-bev256d2-lss.yaml --model pred --checkpoint pretrained/bevfusion-seg.pth --out-dir result/visualize fixed it, but
create_data question
#461

[Ange1ika]

And can you explain how to figure out what is the dimension of the input data, for example, for the lidar-only-det(seg) model, what are the outputs and their dimension ? Is there any way to withdraw it? Sorry for the number of questions

[Ange1ika]

#512 After installation, there was still an error... I started it
torchpack dist-run -np 8 python tools/train.py configs/nuscenes/det/transfusion/secfpn/lidar/voxelnet_0p075.yaml

[Ange1ika]

back to the last message

[Ange1ika]

I've solved almost all those problems.

There was a problem during training - RuntimeError: sigmoid_focal_loss_forward_impl: implementation for device cuda:0 not found. · Issue #228 · mit-han-lab/bevfusion (github.com) ---I have cuda 12.3, and docker 11.3 - most likely, this is the problem, but the solutions presented here did not help me.

[Ange1ika]

I can ask you to describe the dimensions of the main modules of the model for understanding. The article contains fragments, but it is difficult to put them together