The dockerfile seems to be installing a mix of dependencies between Open MMLabs, Nvidia, PyTorch, and TensorRT. I'm struggling to reconcile them.
I try mmcv==1.5.0 but it forces mmcv-full==1.7.0 which fails on the unit testing. And finding the right CUDA version that also connects properly with TensorRT is a struggle, as there's a lot of overlap.
The dockerfile seems to be installing a mix of dependencies between Open MMLabs, Nvidia, PyTorch, and TensorRT. I'm struggling to reconcile them.
I try mmcv==1.5.0 but it forces mmcv-full==1.7.0 which fails on the unit testing. And finding the right CUDA version that also connects properly with TensorRT is a struggle, as there's a lot of overlap.
Is there a way to reconcile these?