Michal Ostyk

Hi, did you manage to try this out @chenzhengnan ? I'm curious about inference speed on a CPU.

@kzky so you completely omitted setting the DISPLAY variable? As far as I understand http://www.open3d.org/docs/latest/docker.html - they claim the opposite.

Just to confirm 1. https://github.com/open-mmlab/mmsegmentation/blob/cbf9af1e3f2eb50b58c8d783008f749ef62f2435/mmseg/apis/utils.py#L29C42-L29C42 2. https://github.com/open-mmlab/mmengine/blob/e43bbb5e03a412ea07f9ef6d0f118586082c2845/mmengine/dataset/base_dataset.py#L20 and the registered transforms in mmengine do not include `LoadImageFromNDArray`  In reality what the destructive result is that we...

Oddly enough `mmdetection` has the support for `LoadImageFromNDArray` but is not a requirement of `mmseg`, `mmcv`, nor `mmengine` https://github.com/open-mmlab/mmdetection/blob/fe3f809a0a514189baf889aa358c498d51ee36cd/mmdet/datasets/transforms/loading.py#L21C38-L21C38