search

DerrickXuNu / v2x-vit

[ECCV2022] Official Implementation of paper "V2X-ViT: Vehicle-to-Everything Cooperative Perception with Vision Transformer"
MIT License
284 stars 32 forks source link

Inference on custom ROS data #9

Closed vcpudding closed 2 years ago

vcpudding commented 2 years ago

Dear author,

I'm trying to make a dataset from my own ROS bag file and use the pretrained V2X-ViT model for inference. But the detection results are never correct. I'm not sure if I get the pcd formats right. Can you suggest the guidelines or scripts to import custom ROS data?

Thanks!

DerrickXuNu commented 2 years ago

Hi,

Did you follow the whole preprocessing pipeline in V2X-ViT? Did you convert your data to opv2v format?

vcpudding commented 2 years ago

I preprocessed the lidar point clouds to make sure the ground plane aligns with the x-y plane. There's only one data source, so I set the extrinsic to identity matrix. My data is not annotated so the ground truth boxes list are left empty. The data folder structure is organized in the opv2v format. Are there other processing pipelines I should follow?

DerrickXuNu commented 2 years ago

can you send some detection visualizations?

DerrickXuNu commented 2 years ago

also, CARLA is using left-hand system. You may want to make sure your data follows that