Tianyuan Zhang

+1, having similar questions

Hi, thanks for testing the fp16 speed and memory consumption. I believe we may see a more significant speedup effect when using GPUs that support FP16 much better, like T4,...

Hi, Thanks for your work. Was the results on DDAD produced by training from scratch using the config setup provided here? https://github.com/TRI-ML/packnet-sfm/blob/master/configs/train_ddad.yaml

> @a1600012888 Yes, that configuration file should work. Thanks!

> Hi, Thanks for your work. > Was the results on DDAD produced by training from scratch using the config setup provided here? https://github.com/TRI-ML/packnet-sfm/blob/master/configs/train_ddad.yaml Hi, for DDAD experiments, Did you...

> after a lot of attemptions,i finally found a robust parameters to re-implement the refined cosine distance method: > 1.we rewrite the cosine distance code. > 2. we set scale...

I know there are several operators in mmdetection3d/mmdetection which might be hard to install. I think checking the similar issues in the repo of mmdetection3d might help. Another way is...

If you want to run the code in nuscenes-mini or any other different dataset, you need to generate a dataset meta file, like here: https://github.com/a1600012888/MUTR3D/blob/main/tools/data_converter/nusc_track.py#L643

It is not used for training more frames. Since I am adding velocity supervisions, the velocity groud truth is from the box position at next frame. That's why I need...

I can't reproduce the results in the paper due to some missing details about the hyper-param in the paper. current results: (linear evaluation protocol on ImageNet). (also on README.md) |Train...