Is deepforest.predict_image() supposed to be restricted to CPU?

[lan13005]

Is the intention that the user must use cpus when predicting with deepforest.predict_image()?

The below code works when device=cpu but crashes when using another device like mps with error slow_conv2d_forward_mps: input(device='cpu') and weight(device=mps:0') must be on the same device.
This makes sense since model is on mps and image is on cpu. Moving image into torch on the same device as the model, e.g. model_device = next(model.parameters()).device in predict._predict_image_() is enough.

On my M5 macbook pro the predicting speed is ~2x faster when using mps.

device = "cpu" # or "mps", ...
model = main.deepforest() # basically lightning module
model.load_model(model_name="weecology/deepforest-tree", revision="main")
model.to(device)
image = ...
model.predict_image(image=image.astype("float32")) # image is required to by numpy

[bw4sz]

I think this is an install issue and specific to the mac chip. I'm going to need a bit more detail on your hardware. No, we use GPU all the time for this function and its all pytorch lightning, nothing in deepforest. Please show the error and we can debug, but this should be specific to your system.