Python API Doesn't Use GPU for pickPoint

[Ahmed Maher]

Hello All,

I am using Metashape Pro's Python API to get the geo-coordinate of each pixel inside the input images (by input images I mean images captured by the physical camera). I reused code already found on this forum. However, when I have a mesh with a large number of faces, I found that chunk.model.pickPoint takes ~ 30 ms per point on a somewhat powerful machine. This makes it totally unusable for a large number of pixels. When I checked the CPU and GPU usage, I found that the GPU is never used even if I explicitly change the GPU mask inside the script while the CPU usage is 100%. As far as I know, the pickPoint function is a ray intersection algorithm that should benefit dramatically from a GPU (the machine has an RTX GPU too). So, how can I make the Python API use the GPU for pickPoint? Is it not currently supported? If it is not supported on Python, is it supported inside the Java API?

[Jordan Pierce]

Finding the same thing sadly. It'd be great if pickPoint was a process ran on a GPU.

For others who are also bummed about the speed of pickPoint, I recommend using a low resolution tiled model instead of mesh or dense point cloud, as they are significantly slower.