fix: is_empty check for filtering

[souhhmm]

Hopefully fixes #1694.

Test code:

import supervision as sv
import numpy as np

CLASSES = [0, 1, 2]

prediction = sv.Detections.empty()
prediction = prediction[np.isin(prediction["class_name"], CLASSES)]

print(prediction)

New output:

Detections(xyxy=array([], shape=(0, 4), dtype=float32), mask=None, confidence=array([], dtype=float32), class_id=array([], dtype=int64), tracker_id=None, data={}, metadata={})

[LinasKo]

I agree with the solution, though I'll move it up.

It happens due to a chain of events:

• getitem for "class_name" returns None.
• np.isin returns an array of shape () with value False.
• __getitem__ is called, numpy uses this array to index into Detections, but since detections are empty, it expands the dimensions instead.

---

# Inside __getitem__
print(self)
print(f"{repr(index)=}")
print(f"{repr(self.xyxy)=}")
print(f"{repr(self.xyxy[index])=}")

Detections(xyxy=array([], shape=(0, 4), dtype=float32), mask=None, confidence=array([], dtype=float32), class_id=array([], dtype=int64), tracker_id=None, data={}, metadata={})
repr(index)='array(False)'
repr(self.xyxy)='array([], shape=(0, 4), dtype=float32)'
repr(self.xyxy[index])='array([], shape=(0, 0, 4), dtype=float32)'

---

Treating the detections as a special case will do for now.

[LinasKo]

@souhhmm, any chance you could:

1. Pull from the branch (I made a small change)
2. Add a unit test in test/detection/test_core.py, either below or inside test_getitem, which checks for this exact situation?

prediction = sv.Detections.empty()
prediction = prediction[np.isin(prediction["class_name"], CLASSES)]

It should return True when result.is_empty() is called.

[souhhmm]

Thanks for your feedback @LinasKo, I've added the test inside test_getitem. Let me know if it needs to changed/fixed.

[LinasKo]

Merged. Thank you @souhhmm! 🤝