Getting Started & Quickstart¶

This guide covers setting up CoreCV and running your first training, inference, and model export pipelines.

Installation¶

CoreCV can be installed using uv (recommended for fast dependency resolution) or standard pip.

Using `uv`¶

uv pip install corecv

Or within a uv project workspace:

uv add corecv

Using `pip`¶

pip install corecv

Install from Source¶

For development or latest features:

git clone https://github.com/liturriago/corecv.git
cd corecv
uv sync

Basic Usage Walkthrough¶

1. Initializing CoreModel¶

The CoreModel class serves as the main entry point to CoreCV:

from corecv.api import CoreModel

# Option A: Initialize directly using a registered backbone string name
model = CoreModel(
    model="resnet18",
    task="classification",
    num_classes=1000,
    input_size=(224, 224),
)

# Option B: Initialize using a raw configuration dictionary
model = CoreModel(
    model={
        "model_name": "resnet50",
        "neck_type": "panet",
        "head_type": "decoupled_anchor_free",
        "neck_channels": 256,
    },
    task="detection",
    num_classes=80,
)

# Option C: Pass a custom PyTorch nn.Module directly
import torch.nn as nn

my_custom_net = nn.Sequential(
    nn.Conv2d(3, 64, kernel_size=3, padding=1),
    nn.BatchNorm2d(64),
    nn.ReLU(),
    nn.AdaptiveAvgPool2d((1, 1)),
    nn.Flatten(),
    nn.Linear(64, 10),
)

model = CoreModel(my_custom_net, task="classification")

2. Model Training¶

Run model training using dictionary configurations, YAML files, or direct keyword arguments:

# Train using keyword arguments
history = model.train(
    data="path/to/dataset",
    epochs=25,
    batch_size=64,
    lr=0.001,
    amp=True,
    target_hardware="edge",
)

3. Model Inference¶

Run inference across image files, directories, or pre-loaded tensors:

# Predict on an image file
results = model.predict("data/sample.jpg", topk=5)

# Output includes confidence scores and labels
for result in results:
    print(f"Predictions: {result.labels}, Scores: {result.scores}")

4. Hardware-Aware Export¶

Export models to edge targets with automatic graph optimization and compile verification:

# Export to ONNX format optimized for edge devices
export_result = model.export(
    format="onnx",
    target_hardware="edge",
    output_path="models/resnet18_edge.onnx",
)

print(f"Exported model saved to: {export_result}")

Next steps: - Learn more about training options in the Training User Guide. - Explore multi-source inference in the Inference User Guide. - Check out edge optimization in the Exporting User Guide.