zzningxp
zzningxp
Furthermore, I think that to test an image by using the pre-trained model on the CPU machine is really in demand. Also, I think that there is only one CPU...
I have GPU machine, and I need to test the forward pregress on the a CPU machine.
From the Readme.md, download it from OneDrive or BaiduYun. Resources Experiment logs: OneDrive, BaiduYun If the automatic "fetch_data" fails, you may manually download resouces from: Pre-complied caffe mex (Windows): OneDrive,...
Should I using the same way like in faster-rcnn to get proposals in a different function using a different caffe prototxt define network and model? Or, just using the current...
The bottleneck of the first block is 1/4 of the channels number after channels concat.
I think it is caused by insufficient training. The network model is too small. Usually, it is suggested that 1) more data, 2) enlarge network capacity.
Solved. By alternating gcc-13 to gcc-11. And this is caused by llama.cpp.
requirement version information: ``` torch=2.6.0=pypi_0 transformers=4.51.1=pypi_0 datasets=3.5.0=pypi_0 dm-tree=0.1.8=pypi_0 pytz=2025.2=pypi_0 fire=0.7.0=pypi_0 wandb=0.19.9=pypi_0 accelerate=1.6.0=pypi_0 tqdm=4.67.1=pypi_0 deepspeed=0.16.5=pypi_0 evaluate=0.4.3=pypi_0 scipy=1.13.1=pypi_0 scikit-learn=1.6.1=pypi_0 promptsource=0.2.3=pypi_0 ```
if you meet same problem please check the PR: https://github.com/NVIDIA-AI-IOT/CUDA-PointPillars/pull/52