search

megvii-research / PETR

[ECCV2022] PETR: Position Embedding Transformation for Multi-View 3D Object Detection & [ICCV2023] PETRv2: A Unified Framework for 3D Perception from Multi-Camera Images
Other
823 stars 125 forks source link

Regarding the use of PETRv2 for semantic segmentation #148

Open LisaShen0509 opened 6 months ago

LisaShen0509 commented 6 months ago

I attempted to obtain BEV semantic segmentation results using the following code with PETRv2: cd /mnt/share/syl/PETR-main chmod 777 tools/dist_test.sh tools/dist_test.sh projects/configs/petrv2/petrv2_BEVseg.py work_dirs/PETRv2_BEVseg/epoch_24.pth 5 --eval segm The console output is shown in the following figure: image The output result is very strange: image Currently, I haven't figured out where the problem lies. Please help me.

wofeifeileni commented 3 months ago

Hi,I would like to know how the segmentation visualization in your second image was obtained and what instructions were used to generate it. Thank you very much. @LisaShen0509

loserhou commented 2 months ago

请问您的问题解决了吗

loserhou commented 2 months ago

你好！运行tools/dist_train.sh/pythondemo/PETR-main/projects/configs/petrv2/petrv2_BEVseg.py 4 --work-dir work_dirs/petrv2_BEVseg/,it has a problem,会报错RuntimeError: Given groups=1, weight of size [128, 256, 3, 3], expected input[1, 1024, 50, 50] to have 256 channels, but got 1024 channels instead,请问你遇到了吗？

Deltaidiots commented 2 months ago

I attempted to obtain BEV semantic segmentation results using the following code with PETRv2: cd /mnt/share/syl/PETR-main chmod 777 tools/dist_test.sh tools/dist_test.sh projects/configs/petrv2/petrv2_BEVseg.py work_dirs/PETRv2_BEVseg/epoch_24.pth 5 --eval segm The console output is shown in the following figure: image The output result is very strange: image Currently, I haven't figured out where the problem lies. Please help me.

my guess would be you are using different checkpoint file as far as i know it should be fcos3d_vovnet_imgbackbone-remapped.pth

Update: i think the threshold value after sigmoid is not taking things into account so we have a problem there

Updated: the threshold seems okay, its not the right checkpoints that i was using and getting the same results: it should be epoch_50.pth after using these checkpoints i am getting almost similar results.