Aymeric Dujardin

@GiorgosBetsos The GTX 1080 doesn't support hardware acceleration for fp16 which is quite a big part for the impressive TensorRT runtime improvement, you could expect about half the runtime in...

I had the same issue. You need to take an older version of darknet, at this commit `80d9bec20f0a44ab07616215c6eadb2d633492fe`. It seems there were some breaking changes after that.

@ferngarc You don't have enough RAM to build on all cores, you can lower the job count by changing this from the dockerfile `make -j$(nproc)` to `make -j2` for instance

Hi, Yes, it's a missing feature, you currently can't (without modifying the plugin). It is planned though. It will probably require PyCUDA.

@frk2 By default the ZED SDK creates a new context, but with the C++ SDK you can pass your own context to the ZED with [InitParameters::sdk_cuda_ctx](https://www.stereolabs.com/developers/documentation/API/v2.4.0/structsl_1_1InitParameters.html#ae684a455a5dc2e48a82b9aadc8c3ba53). As mentioned in the...

Did you install libgflags and glog ? (`sudo apt-get install libgflags2 libgflags-dev libgoogle-glog-dev`)

Hi, Thank you for the detailed report! You're right the current way was too brittle. We'll integrate your fix for the next release

Hi, I'd also love that, but it's currently blocked by Nvidia support of cudagl base images. It's no longer receiving updates and doesn't support Ubuntu 22. More info: https://gitlab.com/nvidia/container-images/cuda/-/issues/147 We...

This is interesting, thanks for your effort! Do you have a sample code where you use the GPU buffer, maybe with Pytorch/TensorRT? Did you notice a significant performance improvement?

Impressive! We need to think about how we could go forward with this. Ideally, I think it would be best to have the cupy dependency optional to limit complexity for...