ADLab-AutoDrive/BEVHeight

custom dataset inference details

Closed this issue · 3 comments

Your work is very amazing. I now have a roadside video. I just want to use the pre-training model to reason and try the effect. What should I do? I only have RGB image data and no camera parameters. Can I infer images without these parameters?

I have the same request, is there a guideline available for how to use the pre-train model to try inference? Thanks.

@Fengzexu For BEVHeight model, camera parameters are necessary during the 2D-3D projection.

@yaobaishen You can utilize the following script to make inferences with a pre-trained model.
[CKPT_PATH] is the path for pre-trained model.
python [EXP_PATH] --ckpt_path [CKPT_PATH] -e -b 8 --gpus 8