DC-Trainer: Modular Experimentation Framework

A flexible, modular, and config-driven framework for deep learning training, designed for combinatorial experimentation with different encoders, sequence models, and decoders for both Classification (ResNet-like) and Generation/OCR (CRNN/Transformer) tasks.

Read the Full Documentation

Features

• Modular Architecture: Plug-and-play Encoders (ConvNext, ResNet), Adapters, Projectors, Sequence Models (Transformer, RNN), and Heads (CTC, Classification).
• Config-Driven: Strictly typed, hierarchical configuration system using YAML.
• Multi-Task: Supports both fixed-label classification and variable-length sequence generation.
• Production Ready: Includes training, evaluation, and inference scripts.

Quick Start

1. Installation

# Clone the repository
git clone https://github.com/Jayden3316/dc-training.git
cd captcha_ocr

# Install dependencies (Python 3.8+)
pip install -r requirements.txt

2. Generate Data

Generate a synthetic dataset using a config file.

python -m cli generate --config-file experiments/dataset_configs/default.yaml --out-dir data/train --dataset-count 1000

3. Run Experiment

Train a model using an experiment configuration.

python -m cli train --config-file experiments/training_configs/generation.yaml

4. Evaluate & Inference

Evaluate on a validation set or run inference on new images.

# Evaluate
python -m cli evaluate --checkpoint checkpoints/best_model.pth --metadata-path data/val/metadata.json

# Inference
python -m cli inference --checkpoint checkpoints/best_model.pth --image-paths test_image.png

---

Configuration & Architecture

The framework uses a pipeline approach: Input -> Encoder -> Adapter -> Projector -> Sequence Model -> Head -> Output

For detailed configuration options and architecture diagrams, please see the Documentation.