Skip to content

CoreCV - Unified Vision Engine

CoreCV Logo

Welcome to CoreCV, a production-ready computer vision engine designed for high-performance model development, seamless training, edge-aware hardware acceleration, and unified multi-format export.

Core Philosophy

CoreCV addresses the fragmentation in vision model deployment by unifying model construction, training, inference, and hardware-targeted export into a cohesive Python API. Built on top of PyTorch, CoreCV bridges the gap between research agility and edge deployment constraints.

  • Unified Facade: Interact with any vision architecture using three primary methods: .train(), .predict(), and .export().
  • Edge-First Engineering: Automatic graph rewriting (TargetRewriter) and zero-VRAM compilation probing (MetaProber) ensure your models run efficiently on constrained edge devices.
  • Type-Safe Modular Architecture: Declarative architecture registration system (CoreRegistry) with signature validation before instantiation.

Feature Matrix

Capability Description CoreCV Component
Unified API Facade pattern consolidating lifecycle management CoreModel
Polymorphic Training Accept configurations via YAML files, dictionaries, or keyword args CoreTrainer
Multi-Source Inference Process single images, directories, numpy arrays, or torch tensors seamlessly CorePredictor
Graph Rewriting Edge-specific hardware optimizations and operator replacements TargetRewriter
Zero-VRAM Validation Validate shape inference and graph exportability on fake meta tensors MetaProber
Multi-Format Export Export models to ONNX and ExecuTorch (.pte) formats CoreExporter
Registry System Generic, type-safe registry system for backbones, necks, heads, and losses CoreRegistry

Unified 3-Step Quickstart

CoreCV simplifies end-to-end vision workflows into three clean steps:

from corecv.api import CoreModel

# 1. Instantiate model directly using a backbone string name
model = CoreModel("resnet18", task="classification", num_classes=10)

# 2. Train with polymorphic arguments
model.train(
    data="path/to/dataset",
    epochs=10,
    lr=1e-3,
    batch_size=32,
    target_hardware="edge",
)

# 3. Predict & Export to edge deployment formats
predictions = model.predict("test_image.jpg", topk=5)
export_paths = model.export(format="onnx", target_hardware="edge")

Next, check out the Quickstart Guide for full installation and setup instructions.