EyeQ

EyeQ is a minimal computer vision inference pakackge. Currently it supports following detectors in onnx runtime. It works with minimal dependencies. It is designed in such a manner to run on edge devices also.

---

Object Detection:

Detector	onnx
yolov5	✅
yolov6	✅
yolov7	✅
yolov8	✅
yolov5u	✅
yoloX	✅
Damo-yolo	✅

---

Instance Segmentation:

Detector Name	onnx
yolov5	✅
yolov7	#TODO
yolov8	✅

---

Multi Object Tracker:

Tracker Name	Integration
SORT	✅
ByteTrack	✅
OcSort	✅
Norfair	-

---

Installation:

Installation can be done via pip using following argument

 pip3 install git+https://github.com/hardikdava/EyeQ.git

---

TODO:

• Docker support
• RestAPI server ✅
• Multi object trackers ✅
• Instance segmentation ✅
• Yolo Dataset loading ✅
• COCO dataset loading
• Object detection evaluation ✅
• Multi object tracker evaluation ✅
• Automatic annotation support using clip, grounding dino and sam
• Introduce SAHI technique

Available APIs:

• Object Detection Inference using ONNX runtime
• Object Detction Evaluation API
• Model serving using RESTAPI using FastAPI based server
• Multi object Tracking for bounding boxes
• Multi object Tracking
• Instance segmentation support
• Data loading for yolo

Note: models are trained using notebooks prepared by roboflow but models are not included with codebase.

References: