Fix RuntimeError in loc_inference.py when using --data_device cpu

[YunSeok-Kang]

Description

This PR fixes a RuntimeError: Expected all tensors to be on the same device that occurs during localization inference (loc_inference.py) when the model is trained with --data_device cpu.

Motivation and Context

When training with --data_device cpu to save VRAM, the ground truth images (qr) remain on the CPU. However, the rendered images (warped_image) are computed on the GPU. This causes a device mismatch in compute_warping_loss when calculating F.mse_loss.

How has this been tested?

I trained the model on the 7Scenes Chess dataset using --data_device cpu and ran loc_inference.py.

• Before fix: Crashed with RuntimeError at warping_loss.py.
• After fix: Localization inference completes successfully without errors.

Changes

• Added a check in warping/warping_loss.py to move the target tensor (qr) to the same device as the input tensor (warped_image) before loss calculation.