[Compatibility Issue] np.bool removed in NumPy 1.24 — use bool instead

[xyf5432]

Affected File

lib/utils/data_utils.py:410-411

Current Code

def inter_from_mask(pred, gt):
    pred = pred.astype(np.bool)
    gt = gt.astype(np.bool)
    intersection = np.logical_and(gt, pred).sum()
    return intersection

Root Cause

np.bool was deprecated in NumPy 1.20 and fully removed in NumPy 1.24 (2022-12). In NumPy >= 1.24, using np.bool raises:

AttributeError: module 'numpy' has no attribute 'bool'

np.bool was an alias for Python's built-in bool.

Impact

• Severity: High — hard crash at runtime on NumPy >= 1.24
• Affects the inter_from_mask utility function used for intersection computation between prediction and ground-truth masks

Dependency Chain

Although requirements.txt does not list numpy directly, it is pulled in transitively by every declared dependency:

Package	Requires numpy
scikit-image==0.19.0	numpy>=1.24
opencv-python	numpy>=2 (Python 3.9+)
kornia	numpy<3 (dev extra)
imgaug	numpy>=1.15
lpips	numpy>=1.14.3
imageio==2.27.0	numpy
plyfile	numpy>=2.0
tensorboardX	numpy

The resolver takes the tightest constraints: numpy>=2.0 (from plyfile and opencv-python). Any numpy >= 2.0 has already removed np.bool, so this issue is guaranteed to trigger.

Solution

# Before:
pred = pred.astype(np.bool)
gt = gt.astype(np.bool)

# After:
pred = pred.astype(bool)
gt = gt.astype(bool)

Python's built-in bool is the canonical replacement. np.bool_ can also be used if NumPy scalar type is explicitly required.

References

• NumPy 1.24 release notes — np.bool removed
• NumPy 1.20 deprecation: https://numpy.org/doc/stable/release/1.20.0-notes.html#deprecations
• Similar fix in another downstream project: ultralytics#22930