anycv 0.2.2

A simple, unified computer vision inference toolkit. One-liner API for common CV tasks.

anycv

Computer vision in one line of code

A simple, unified computer vision inference toolkit. One-liner API for object detection, image classification, and semantic segmentation using ONNX Runtime.

Runs completely offline after first model download. Models are fetched from HuggingFace Hub once and cached locally in ~/.cache/anycv/. No cloud APIs required.

Built by Viet-Anh Nguyen at NRL.ai.

Installation

pip install anycv

For GPU support:

pip install anycv[gpu]

Quick Start

Object Detection

import anycv

# Runs YOLOv8n object detection via ONNX Runtime
# Model auto-downloads from HuggingFace Hub (~6MB), cached in ~/.cache/anycv/
# Returns Detection(bbox, label, confidence) for each detected object
detections = anycv.detect("photo.jpg")
for det in detections:
    print(f"{det.label}: {det.confidence:.2f} at {det.bbox}")

Image Classification

# Runs MobileNetV2 classification via ONNX Runtime (~14MB model)
# Preprocesses image to 224x224, normalizes with ImageNet mean/std
# Returns top-k ClassificationResult(label, confidence) from 1000 ImageNet classes
classes = anycv.classify("photo.jpg")
print(f"Top prediction: {classes[0].label} ({classes[0].confidence:.2%})")

Semantic Segmentation

# Runs DeepLabV3 segmentation via ONNX Runtime
# Produces a per-pixel class map over 21 Pascal VOC categories
# Returns SegmentationResult with .labels (classes found) and .mask (H, W array)
seg = anycv.segment("photo.jpg")
print(f"Classes found: {seg.labels}")
# Access the pixel-level mask
mask = seg.mask  # (H, W) array of class IDs

Batch Processing

# Process multiple images at once
detections = anycv.detect(["img1.jpg", "img2.jpg", "img3.jpg"])

Flexible Image Input

import numpy as np
from PIL import Image

# All of these work:
anycv.detect("path/to/image.jpg")           # File path
anycv.detect(Path("image.png"))              # pathlib.Path
anycv.detect("https://example.com/img.jpg")  # URL
anycv.detect(np.zeros((480, 640, 3)))        # NumPy array
anycv.detect(Image.open("photo.jpg"))        # PIL Image

Available Models

Model	Task	Description
yolov8n	Detection	YOLOv8 Nano -- fast, 80 COCO classes
yolov8s	Detection	YOLOv8 Small -- balanced, 80 COCO classes
mobilenetv2	Classification	MobileNetV2 -- lightweight, 1000 ImageNet classes
resnet50	Classification	ResNet-50 -- accurate, 1000 ImageNet classes
deeplabv3	Segmentation	DeepLabV3 -- 21 Pascal VOC classes

# List all available models
for model in anycv.list_models():
    print(f"{model.name:15s} {model.task:15s} {model.description}")

# Filter by task
det_models = anycv.list_models(task="detection")

Advanced Usage

Load and Reuse Models

model = anycv.load_model("yolov8n")
results1 = model(["img1.jpg", "img2.jpg"])
results2 = model("img3.jpg")

Custom Thresholds

detections = anycv.detect("photo.jpg", conf_threshold=0.5, iou_threshold=0.6)

Drawing Results

from anycv import load_image, draw_boxes

image = load_image("photo.jpg")
detections = anycv.detect(image)

boxes = [d.bbox for d in detections]
labels = [f"{d.label} {d.confidence:.0%}" for d in detections]
annotated = draw_boxes(image, boxes, labels=labels)

Local-First / Edge AI

This package is designed to work completely offline. After initial model download, no internet connection is required. All inference runs locally via ONNX Runtime on CPU, GPU, or edge devices.

# Pre-download all models for offline use
python -m anycv download

# Download only detection models
python -m anycv download --task detection

import anycv

# Pre-download everything
anycv.download_models()

# Download only classification models
anycv.download_models(task="classification")

Benchmarks

Model	Input Size	CPU (ms)	GPU (ms)
yolov8n	640x640	TBD	TBD
yolov8s	640x640	TBD	TBD
mobilenetv2	224x224	TBD	TBD
resnet50	224x224	TBD	TBD
deeplabv3	513x513	TBD	TBD

Development

git clone https://github.com/vietanhdev/anycv.git
cd anycv
pip install -e ".[dev]"
pytest tests/ -v

License

MIT License. See LICENSE for details.

Links

• GitHub: github.com/vietanhdev/anycv
• NRL.ai: www.nrl.ai
• Author: Viet-Anh Nguyen