Repository navigation
Leveraging additional information from nuscenes #38
Description
Hi there, we're considering if the agent could benefit from more than just keyframe images to generate the right answers. Is that possible to use some extra infomations from nuscenes in this task, like continuous frame images or radar points, which can be used to obtain more precise velocity information?
Activity
ChonghaoSima commentedon Mar 14, 2024
you are free to do so
ChonghaoSima commentedon Mar 14, 2024
you can also consider other modality as long as it's provided by nusc dataset
dongyang2011 commentedon Mar 14, 2024
So the full nuscenes dataset will be provided at challenge/llama_adapter_v2_multimodal7B/data/nuscene, right? And I guess the annotations, such as map expansion, lidar seg... will not be available.
ChonghaoSima commentedon Mar 14, 2024
no, we won't provide the "full" nuscenes dataset here. you can refer to bevformer or other repo to download the full dataset.
no, those annotation (bbox, map, lidar seg) should not be used when inference.
dongyang2011 commentedon Mar 14, 2024
Do you mean that you will not provide the full nuscenes in this repo, but on the test server we can somehow access to the full nuscenes dataset (without any annotation)?
ChonghaoSima commentedon Mar 14, 2024
not sure what do you mean by "on the test server we can somehow access to the full nuscenes dataset".
we do not restrict the input modality and history frames for the inference of the model, while we don't allow using any human-labelled annotation as well as nusc provided ground truth annotation (including but not limited to bbox, map, lidar seg).
dongyang2011 commentedon Mar 14, 2024
Thanks !
dongyang2011 commentedon Mar 14, 2024
Is it opposite of the Huggingface homepage - Specific Rules?
ChonghaoSima commentedon Mar 14, 2024
typo there. will fix soon.