Lue Fan
Lue Fan
Thanks for the feedback. So, can you reproduce the performance with your re-trained models? @yrt1993
Thanks for your kind feedback. Glad to see that. Good luck with your exploration!
Sorry, we are too busy now. We are planning to release these in the middle of March.
The Argoverse model is released now. See the instructions: https://github.com/tusen-ai/SST/blob/main/instructions.md @rentoumacsj @foolhard The nuScenes config will be updated later since it is related to our new work.
@rentoumacsj We use single-stage model for Argo2 by default because the IoU-based assignment of group correction is not suitable for the distance-based evaluation metric of Argo2.
In fact, numpy=1.21.0 is ok in our repo. You can install NumPy 1.21.0 after installing mmdet3d. Some warnings related to data type might appear after installing NumPy 1.21.0, you could...
It seems a bug of av2-api, try the following instruction: In L681 of `av2/map/map_api.py`, change `log_map_dirpath / vector_data_frame` to `vector_data_frame` L681 might be not accurate if you install a different...
[data_augment.txt](https://github.com/tusen-ai/RangeDet/files/9926722/data_augment.txt) This is our internal aug code, but may be not completely compatible with current code. FYI.
We limit the max number of points in each box to reduce the overhead, in case of an object containing too many points. In the extreme case of Waymo Open...
Sorry for late reply. Do you want to use the pretrained CTRL or train a new CTRL model with your own data?