Detections _get_item_() does not handle index arrays of booleans correctly for data dictionary in Detections object.

[rolson24]

Search before asking

• I have searched the Supervision issues and found no similar bug report.

Bug

When I tried to use boolean array indexing to get specific detections from a Detections object, the data field sometimes returns the wrong values (everything is the second value) and sometimes it just doesn't work. See output of code:

valid detections data: {'random': array([0.69156142, 0.43520246, 0.08080701, 0.32277943]), 'Dataset': ['Roboflow 100', 'Roboflow 100', 'Roboflow 100', 'Roboflow 100']}
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
[<ipython-input-6-6cae6894f883>](https://localhost:8080/#) in <cell line: 31>()
     29 # This does not run
     30 detection_inds = np.asarray([True, False, True, False], dtype = bool)
---> 31 print(f"valid detections data: {detections[detection_inds].data}")

4 frames
[/usr/local/lib/python3.10/dist-packages/supervision/detection/core.py](https://localhost:8080/#) in __getitem__(self, index)
    928         if isinstance(index, int):
    929             index = [index]
--> 930         return Detections(
    931             xyxy=self.xyxy[index],
    932             mask=self.mask[index] if self.mask is not None else None,

/usr/local/lib/python3.10/dist-packages/supervision/detection/core.py in __init__(self, xyxy, mask, confidence, class_id, tracker_id, data)

[/usr/local/lib/python3.10/dist-packages/supervision/detection/core.py](https://localhost:8080/#) in __post_init__(self)
     88 
     89     def __post_init__(self):
---> 90         validate_detections_fields(
     91             xyxy=self.xyxy,
     92             mask=self.mask,

[/usr/local/lib/python3.10/dist-packages/supervision/detection/utils.py](https://localhost:8080/#) in validate_detections_fields(xyxy, mask, class_id, confidence, tracker_id, data)
    750     validate_confidence(confidence, n)
    751     validate_tracker_id(tracker_id, n)
--> 752     validate_data(data, n)
    753 
    754 

[/usr/local/lib/python3.10/dist-packages/supervision/detection/utils.py](https://localhost:8080/#) in validate_data(data, n)
    724         if isinstance(value, list):
    725             if len(value) != n:
--> 726                 raise ValueError(f"Length of list for key '{key}' must be {n}")
    727         elif isinstance(value, np.ndarray):
    728             if value.ndim == 1 and value.shape[0] != n:

ValueError: Length of list for key 'Dataset' must be 2

Environment

• Supervision 0.18.0 - google colab

Minimal Reproducible Example

https://colab.research.google.com/drive/1C-wlttD-awePYPTCCPmMfIEtME9o3dN7?usp=sharing

import numpy as np

import supervision as sv
import random


bboxes = np.asarray(([1,50,50,100], [40,30,70,80], [3,90,150,120], [95,50,140,100]))
confidence = np.asarray([.8, .75, .69, .99])
class_id = np.asarray([2, 4, 2, 29])
data = {"random": np.asarray([random.random(), random.random(), random.random(), random.random()]), "Dataset": ["KITTI", "Roboflow 100", "nuScenes", "COCO"]}
detections = sv.Detections.empty()
detections.xyxy = bboxes
detections.confidence = confidence
detections.class_id = class_id
detections.data = data

""" example might be filtering out detections that have a certain class by doing
    this:
      dog_class = 4
      detections = detections[detection.class_id != dog_class]

    (detections.class_id != dog_class with return a np.ndarray of type bool)
"""

# This runs but is incorrect
detection_inds = np.asarray([True, True, True, True], dtype = bool)
print(f"valid detections data: {detections[detection_inds].data}")

# This does not run
detection_inds = np.asarray([True, False, True, False], dtype = bool)
print(f"valid detections data: {detections[detection_inds].data}")

Additional

I am pretty sure this is because of how the get_data_item() function in the Detections.utils file is handling indexes that are numpy arrays. I think it's a pretty simple fix, and I can submit a PR.

Are you willing to submit a PR?

• Yes I'd like to help by submitting a PR!

[SkalskiP]

Hi @rolson24! 👋🏻 good catch! I conducted my own test, and its results confirm your conclusions.

import numpy as np
import supervision as sv

xyxy = np.array([
    [100, 100, 110, 110],
    [200, 200, 220, 220],
    [300, 300, 330, 330],
    [400, 400, 440, 440],
    [500, 500, 550, 550]
], dtype=float)

confidence = np.array([
    0.1, 0.2, 0.3, 0.4, 0.5
], dtype=float)

class_id = np.array([
    1, 2, 3, 4, 5
], dtype=int)

letter = [
    'a', 'b', 'c', 'd', 'e'
]

detections = sv.Detections(
    xyxy=xyxy,
    class_id=class_id,
    confidence=confidence,
    data={
        "letter": letter
    }
)

detections[[1, 2]]

# Detections(
#     xyxy=array([
#         [200., 200., 220., 220.],
#         [300., 300., 330., 330.]]), 
#     mask=None, 
#     confidence=array([0.2, 0.3]), 
#     class_id=array([2, 3]), 
#     tracker_id=None, 
#     data={'letter': ['b', 'c']}
# )
# OK.

detections[np.array([1, 2])]

# Detections(
#     xyxy=array([
#         [200., 200., 220., 220.],
#         [300., 300., 330., 330.]]), 
#     mask=None, 
#     confidence=array([0.2, 0.3]), 
#     class_id=array([2, 3]), 
#     tracker_id=None, 
#     data={'letter': ['b', 'c']}
# )
# OK.

detections[np.array([1, 1, 1, 1, 1])]

# Detections(
#     xyxy=array([
#         [200., 200., 220., 220.],
#         [200., 200., 220., 220.],
#         [200., 200., 220., 220.],
#         [200., 200., 220., 220.],
#         [200., 200., 220., 220.]]), 
#     mask=None, 
#     confidence=array([0.2, 0.2, 0.2, 0.2, 0.2]), 
#     class_id=array([2, 2, 2, 2, 2]), 
#     tracker_id=None, 
#     data={'letter': ['b', 'b', 'b', 'b', 'b']}
# )
# OK.

detections[np.array([True, True, True, True, True], dtype=bool)]

# Detections(
#     xyxy=array([
#         [100., 100., 110., 110.],
#         [200., 200., 220., 220.],
#         [300., 300., 330., 330.],
#         [400., 400., 440., 440.],
#         [500., 500., 550., 550.]]), 
#     mask=None, 
#     confidence=array([0.1, 0.2, 0.3, 0.4, 0.5]), 
#     class_id=array([1, 2, 3, 4, 5]), 
#     tracker_id=None, 
#     data={'letter': ['b', 'b', 'b', 'b', 'b']}
# )
# WRONG DATA.

detections[np.array([False, False, False, False, False], dtype=bool)]

# ---------------------------------------------------------------------------
# ValueError                                Traceback (most recent call last)
# [<ipython-input-26-634699880563>](https://localhost:8080/#) in <cell line: 1>()
# ----> 1 detections[np.array([False, False, False, False, False], dtype=bool)]
# 
# 4 frames
# [/usr/local/lib/python3.10/dist-packages/supervision/detection/utils.py](https://localhost:8080/#) in validate_data(data, n)
#     724         if isinstance(value, list):
#     725             if len(value) != n:
# --> 726                 raise ValueError(f"Length of list for key '{key}' must be {n}")
#     727         elif isinstance(value, np.ndarray):
#     728             if value.ndim == 1 and value.shape[0] != n:
# 
# ValueError: Length of list for key 'letter' must be 0

[SkalskiP]

It was fixed with #1062. Merging!

[rolson24]

Great! Thanks!