more details about image-text pair data

[annopackage]

Hi,
Thanks for your great work.

I have several questions about the data and method:

1. I am curious about the pipeline about generation of text list about HD map. Can you share more details about how to get text for multi-view and bev images? Are those information from a pretrained mulit-modal model or rules based on hd map?
2. Are the visual encoder the same for multi-view images and bev cloud images? The encoder in the paper seems different, but in the inference code https://github.com/LLVM-AD/MAPLM/blob/main/baseline/evaluation/inference.py#L72C29-L72C44, the image processors are the same.

[IrohXu]

Thanks for your question. For the first question, they are from a pretrained mulit-modal model based on Tencent's THMA. This type of data is restricted by the company policy and we can not release them currently. For the second question, we are using different encoders for RGB inputs and BEV input. The inference code is the default version of the baseline. We will release the weight for BEV encoder soon.

We are also working on the v2 version of the dataset and annotations. It will be released here: https://huggingface.co/datasets/LLVM-AD/maplm_v2