Implementation:Neuml Txtai Object Detection
| Knowledge Sources | |
|---|---|
| Domains | Machine Learning, Computer Vision, Object Detection, Image Classification, Transformers |
| Last Updated | 2026-02-10 01:00 GMT |
Overview
Concrete tool for applying object detection and image classification models to images provided by txtai.
Description
Objects extends HFPipeline and wraps either the Hugging Face object-detection or image-classification pipeline depending on the classification flag. It detects objects in images and returns (label, score) tuples sorted by descending confidence score. Results are filtered by a configurable confidence threshold (default 0.9) and deduplicated by label (keeping the highest-scoring instance of each label). The flatten parameter returns just label strings instead of tuples. The pipeline requires the PIL (Pillow) library.
Usage
Use Objects when you need to detect and label objects in images or classify images into categories. Object detection mode returns labeled bounding box results, while classification mode returns image-level labels. This is useful for content tagging, visual search, and image understanding workflows.
Code Reference
Source Location
- Repository: Neuml_Txtai
- File:
src/python/txtai/pipeline/image/objects.py
Signature
class Objects(HFPipeline):
def __init__(self, path=None, quantize=False, gpu=True, model=None, classification=False, threshold=0.9, **kwargs)
def __call__(self, images, flatten=False, workers=0)
Import
from txtai.pipeline.image.objects import Objects
I/O Contract
Inputs
| Name | Type | Required | Description |
|---|---|---|---|
| path | str | No | Model path; accepts Hugging Face model hub id or local path. Uses default model for task if not provided. |
| quantize | bool | No | If True, quantizes the model to int8 (CPU only). Defaults to False. |
| gpu | bool or int | No | True/False to enable GPU, or a specific GPU device id. Defaults to True. |
| classification | bool | No | If True, uses image classification instead of object detection. Defaults to False. |
| threshold | float | No | Minimum confidence score for including results. Defaults to 0.9. |
| images | str, PIL.Image, or list | Yes (call) | A single image (file path string or PIL Image) or a list of images. |
| flatten | bool | No (call) | If True, returns just label strings instead of (label, score) tuples. Defaults to False. |
| workers | int | No (call) | Number of concurrent workers for data processing. Defaults to 0. |
Outputs
| Name | Type | Description |
|---|---|---|
| result | list | List of (label, score) tuples sorted by descending score and deduplicated by label. If flatten=True, returns a list of label strings. For single image input, returns a 1D list; for list input, returns a 2D list. |
Usage Examples
from txtai.pipeline.image.objects import Objects
# Create an object detection pipeline
detector = Objects("facebook/detr-resnet-50", gpu=True)
# Detect objects in a single image
results = detector("street_scene.jpg")
# Returns: [("car", 0.98), ("person", 0.95), ("traffic light", 0.92)]
# Flatten to just labels
results = detector("street_scene.jpg", flatten=True)
# Returns: ["car", "person", "traffic light"]
# Detect objects in multiple images
results = detector(["image1.jpg", "image2.jpg"])
# Use image classification mode
classifier = Objects(classification=True, threshold=0.5)
results = classifier("cat.jpg")
# Returns: [("tabby cat", 0.95), ("domestic cat", 0.88)]
# Lower the threshold for more results
detector = Objects(threshold=0.5)
results = detector("scene.jpg")