code_explain

My server has four NVIDIA 4090 GPUs. Single-card training doesn't throw any errors, but when the batch size is changed to 2 for single-card training, it throws an error after...

ch2 报错

2

报错信息如下 /home/xxx/slam_in_autonomous_driving/src/tools/pointcloud_convert/velodyne_convertor.cc:5:10: fatal error: tools/pointcloud_convert/velodyne_convertor.h: 没有那个文件或目录 5 | #include "tools/pointcloud_convert/velodyne_convertor.h" | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ compilation terminated. 这个文件是存在的，为什么会找不到呢？不太清楚应该修改那个地方了 我是在src/建立的build文件夹 cd build cmake ..没报错 make -j2 报错误上 另外把src文件夹下的CMakeLists中的ch3~ch10都注释掉了。

The results of the two inferences are inconsistent.

3

Hello, I have encountered an issue during inference. With the same configuration, the prediction results from two inferences are inconsistent, with a small portion of the data having inconsistent labels...

If I only want to use the provided model and collect my own data for inference, how should I modify the code?