learnt the physical structure of GPU, and logical grid, block and threads of GPU. These can help us know about the cuda program writing.

know about the cuda programming, like __global__ function(running on GPU) and __device__ function (running on CPU).

I learnt CUDA programming based on c++. In 3D point cloud processing, some IOU3D, NMS algorithms do not have engaged python package. So, in the consideration of the inference speed, some algorithms should be realized by C++, and make self-made python packages for whole project.

In this week, I uesd most of the time to research about the PointRCNN program skills. And I basically gave up the plan for fusing lidar and camera. Later time, I decided to research lidar-only 3D detection frameworks.

I finished the ICISIP2022 paper this week.

I still cannot figure out the principle of fusion works after I ran the open source program. Finally, I decided to write a summary for ICISIP2022, also help me to understand more details about the point clouds processing methods. Until now, I have write 4 pages. And I can finish this paper in … Continue Reading ››

1. About my research, I try to read the papers and run open source works. (like transfusion, SFD, PV-RCNN) But still not getting my own ideas.
2. I finished the research plan in this year, and uploaded to the gitlab.
3. I still did not finish my icisip paper.

I read about the sensor fusion papers. And I get some new knowledge to make multi-fusion frameworks. The fusion work can be divided into two parts.

1. One is to fuse camera and lidar in 2D space. This fusion in 2D space is an normal fusion method in the past … Continue Reading ››

７月３日、日本語テストがありますがら、今週の大分の時間に日本語お練習しました。

1. 最近は、カメラとLIDARのキャリブレーションに取り組んでいます。 対応するコードが見つかりました。 現在、シミュレーションに取り組んでいます。
2. 3Dターゲット検出のためのマルチセンサーフュージョンはあまり進歩していない。 まだ、文献を読んでいる段階。