Fix: Tensor Type Issue for cv2.boxPoints() in gen_bbx() by Explicit Float Conversion

[dongminlee94]

Summary

This pull request resolves the parsing error that occurs when OpenCV’s cv2.boxPoints() receives PyTorch tensor scalars instead of Python floats. By explicitly converting the tensor values (x, y, w, h, theta) to float, we ensure that the function receives the correct data type for constructing a valid rotated rectangle, preventing the Can't parse 'box' center point. Sequence item with index 0 has a wrong type error.

Changes

In the gen_bbx() function, replaced the direct usage of PyTorch tensors with explicit float casts.

Specifically, changed:

box = ((x, y), (w, h), theta)

to

box = ((float(x), float(y)), (float(w), float(h)), float(theta))

This ensures that each value is a Python float before being passed to cv2.boxPoints().

Benefits

• Eliminates the Can't parse 'box' center point error by providing valid (center, size, angle) inputs to cv2.boxPoints().
• Allows the shadow generation (and any other functionality relying on gen_bbx()) to proceed without runtime failures.
• Improves compatibility between PyTorch and OpenCV in scenarios where tensor scalars must be interpreted as floats.

Testing

• Verified by running the existing test_shadow_generation.py within libcom/tests/.
• After applying the float conversion, the error no longer occurs, and the shadow generation results are produced successfully.