HopeJW comments

Results 128 comments of


                                            HopeJW

How should I export the data of camera_bevfeat of type const nvtype::half*, because I want to visualize the BEV results of Camera. Thank you very much.

You can use the save function in class::Tensor. For example([link](https://github.com/NVIDIA-AI-IOT/Lidar_AI_Solution/blob/d4f4ba54c7a0313b991284b1c22b1b9dfa66e29f/CUDA-BEVFusion/src/common/tensor.hpp#L136)): ```c++ nv::Tensor::from_data_reference(camera_bevfeat, {1, 2, 3 /* the shape of camera_bevfeat */}, nv::Tensor::DataType::Float16, true).save("camera_bevfeat.tensor"); ``` And then, you can load it...

有没有详细一点的web_demo推理部署流程？

你们的demo的使用说明，不能用垃圾来形容，简直无可救药！！！

which model should I use?

Would you like to deploy a **lidar only** detector? No camera, right?

which model should I use?

6 cameras are not necessary. You can still change to 3 cameras easily.

which model should I use?

Yeah, you are right.

How to run the nuscenes dataset with CUDA-BEVFusion?

Please refer to https://github.com/NVIDIA-AI-IOT/Lidar_AI_Solution/blob/master/CUDA-BEVFusion/qat/test-mAP-for-cuda.py

Spconv Engine Implementation

I'm sorry, but unfortunately we have no plan to public the libspconv.so. The libspconv.so in our repository is independent of other implementations. Therefore, you may not find the class in...

Spconv Engine Implementation

First of all, I want to clarify that there is no leading or different acceleration technology here. The following might be worth considering: 1. Inference on FP32, FP16 or INT8?...

how to train and use myself data

It would help if you referred to the [BEVFusion repository](https://github.com/mit-han-lab/bevfusion?tab=readme-ov-file#data-preparation).

Datatype shows "Assert failed 💀" after reconfig layer precision

因为要确保add算子的左右两侧是相同数据类型的。确认一下输入给add的两个input是不是都是int8或者fp16.