Gao, Ruiyuan
Gao, Ruiyuan
We follow the implementation of ControlNet, where the control signal (map) is first added with x (noisy latent) and then goes through the copied encoder block. The whole file you...
@lyx3911 我这边单机测试几乎没有提升，能说下具体什么场景下有提速吗？是否需要 `overlap_allgather` 参数设置成 `True`？谢谢
No, it should be 8 in the latest version. We use raw_box for objects, so there is no object-related layers in the map.
Please consider building them like real trajectories such as those from the validation set. It may not work on arbitrary parameters.
Like I said, please do your best to mimic a real trajectory and see what will happen. I think you should be able to find real trajectory as examples for...
Please understand I can only offer help to basic usage of our models, and may not have time to look into details of your use cases. Since this is an...
Please check the webui first. We have screenshots in the readme to change it to "rainy". I do not think the model can generate "foggy" because our training data does...
nuScenes. BTW, I suggest you to take a look at our paper, which contains much information about our work.
We use 32G V100. I did not try others.
@jspark5638 Sorry I cannot remember the exact number but inference does not take much.