Retina: Object Detection Skills

SUMMARY

Retina is a module in the Telekinesis SDK containing skills for 2D object detection from images.

It provides detection capabilities using classical computer vision techniques, deep learning models, and modern foundation models, enabling reliable visual perception across a wide range of robotics applications.

When to Use Retina?

Use Retina when your robotics application relies on 2D visual detection from cameras, such as:

• Object detection for vision-guided manipulation
• Visual target identification for pick-and-place pipelines
• Perception for mobile robot navigation and obstacle awareness
• Visual inspection and monitoring in robotic systems
• Camera-based input for Physical AI agents

What Does Retina Provide?

Retina includes a collection of modular skills for:

• 2D object detection using classical vision algorithms
• Deep learning–based object detectors
• Foundation model–based and prompt-driven detection
• Image preprocessing and feature extraction for detection pipelines

How to Use Retina?

To use the skills from Retina, simply use:

from telekinesis import retina

Here is a minimal example:

from telekinesis import retina
from datatypes import datatypes, io

# Load an image
image = io.load_image(filepath=datatypes.String("path/to/image.png"))

# Detect circles using classic Hough transform
annotations = retina.detect_circle_using_classic_hough(
    image=image,
    inverse_resolution_ratio=1,
    min_distance=50,
    min_radius=40,
    max_radius=60,
    canny_detector_upper_threshold=300,
    accumulator_threshold=30,
)

# Access the annotations
annotations = annotations.to_list()

Overview of Skills

SDK Call	Description
detect_circle_using_classic_hough	Detects circles using the classic Hough transform.
detect_contours	Detects object outlines using a contour-based detector.
detect_objects_using_rfdetr	Detects objects in images using the RF-DETR model.
detect_objects_using_yolox	Detects objects in images using the YOLOX model.
detect_objects_using_grounding_dino	Detects objects in images using Grounding DINO (zero-shot).
detect_objects_using_qwen	Detects objects in images using Qwen vision-language model.