Examples¶

The examples/ directory contains runnable scripts demonstrating AutoTimm features.

Examples Organization¶

graph LR
    A[<b>AutoTimm<br/>Examples</b>] --> B[<b>Getting Started</b><br/>Classification,<br/>data loading]
    A --> C[<b>Computer Vision</b><br/>Detection,<br/>segmentation]
    A --> D[<b>HuggingFace</b><br/>Hub integration,<br/>model sharing]
    A --> E[<b>Training</b><br/>HPO, multi-GPU,<br/>auto-tuning]
    A --> F[<b>Understanding</b><br/>GradCAM, attention,<br/>feature maps]
    A --> G[<b>CLI</b><br/>YAML configs,<br/>command line]

    style A fill:#1565C0,stroke:#0D47A1
    style B fill:#1976D2,stroke:#1565C0
    style C fill:#1565C0,stroke:#0D47A1
    style D fill:#1976D2,stroke:#1565C0
    style E fill:#1565C0,stroke:#0D47A1
    style F fill:#1976D2,stroke:#1565C0
    style G fill:#1565C0,stroke:#0D47A1

Quick Reference¶

Getting Started¶

Classification - Start with CIFAR-10 and custom datasets
HuggingFace Hub Basics - Load and use HF Hub models

Computer Vision Tasks¶

Object Detection - FCOS, YOLOX, RT-DETR, and transformers
Semantic Segmentation - DeepLabV3+ and FCN
Instance Segmentation - Mask R-CNN style

HuggingFace Advanced¶

Model Interpretation - GradCAM, attention visualization, metrics
Transfer Learning - LLRD, progressive unfreezing
Ensemble & Distillation - Model ensembles and knowledge distillation
Deployment - ONNX, quantization, serving
Custom Data - Advanced augmentation and data handling
Hyperparameter Tuning - Optuna integration

Training & Optimization¶

Training Utilities - Auto-tuning, multi-GPU, presets
Data Handling - Balanced sampling, augmentation
Logging & Metrics - TensorBoard, MLflow, evaluation

CLI¶

CLI Examples - Train from YAML configs on the command line

Model Understanding¶

Interpretation Methods - Comprehensive interpretation toolkit
Backbone Utilities - Discover and compare models

All Examples by Category¶

Getting Started¶

Script	Description
`getting_started/classify_cifar10.py`	ResNet-18 on CIFAR-10 with MetricManager
`getting_started/classify_custom_folder.py`	Train on custom ImageFolder dataset with W&B
`getting_started/vit_finetuning.py`	Two-phase ViT fine-tuning

Computer Vision Tasks¶

Object Detection:

Script	Description
`computer_vision/object_detection_coco.py`	FCOS-style object detection on COCO
`computer_vision/object_detection_yolox.py`	YOLOX object detection
`computer_vision/object_detection_rtdetr.py`	RT-DETR transformer detection
`computer_vision/object_detection_transformers.py`	Vision Transformers for detection
`computer_vision/explore_yolox_models.py`	Explore YOLOX model variants
`computer_vision/yolox_official.py`	Official YOLOX implementation

Segmentation:

Script	Description
`computer_vision/semantic_segmentation.py`	DeepLabV3+ semantic segmentation
`computer_vision/instance_segmentation.py`	Mask R-CNN style instance segmentation

HuggingFace Hub Integration¶

Basic Integration:

Script	Description
`huggingface/huggingface_hub_models.py`	Introduction to HF Hub
`huggingface/hf_hub_classification.py`	Classification with HF Hub
`huggingface/hf_hub_segmentation.py`	Segmentation with HF Hub
`huggingface/hf_hub_object_detection.py`	Detection with HF Hub
`huggingface/hf_hub_instance_segmentation.py`	Instance segmentation with HF Hub
`huggingface/hf_hub_advanced.py`	Advanced HF Hub features
`huggingface/hf_hub_lightning_integration.py`	Lightning compatibility
`huggingface/hf_direct_models_lightning.py`	Direct HF Transformers models

Advanced HF Hub:

Script	Description
`huggingface/hf_interpretation.py`	Model interpretation (GradCAM, attention, metrics)
`huggingface/hf_transfer_learning.py`	LLRD, progressive unfreezing
`huggingface/hf_ensemble.py`	Ensembles and knowledge distillation
`huggingface/hf_deployment.py`	ONNX export, quantization, serving

Data & Augmentation¶

Script	Description
`data_training/balanced_sampling.py`	Weighted sampling for imbalanced data
`data_training/albumentations_cifar10.py`	Albumentations strong augmentation
`data_training/albumentations_custom_folder.py`	Custom albumentations pipeline
`data_training/multilabel_classification.py`	Multi-label classification with CSV data
`data_training/csv_classification.py`	Classification from CSV files
`data_training/csv_detection.py`	Object detection from CSV annotations
`data_training/csv_segmentation.py`	Semantic segmentation from CSV
`data_training/csv_instance_segmentation.py`	Instance segmentation from CSV
`data_training/hf_custom_data.py`	Advanced augmentation, multi-label, validation

Training & Optimization¶

Script	Description
`data_training/auto_tuning.py`	Automatic LR and batch size finding
`data_training/multi_gpu_training.py`	Multi-GPU and distributed training
`data_training/preset_manager.py`	Training presets and configurations
`data_training/performance_optimization_demo.py`	Performance optimization techniques
`data_training/hf_hyperparameter_tuning.py`	Optuna hyperparameter optimization

Model Interpretation¶

Script	Description
`interpretation/interpretation_demo.py`	Basic interpretation and visualization
`interpretation/interpretation_metrics_demo.py`	Interpretation with metrics analysis
`interpretation/interpretation_phase2_demo.py`	Advanced techniques (Phase 2)
`interpretation/interpretation_phase3_demo.py`	Advanced techniques (Phase 3)
`interpretation/interactive_visualization_demo.py`	Interactive Plotly visualizations
`interpretation/comprehensive_interpretation_tutorial.ipynb`	Comprehensive tutorial (Notebook)

Logging & Experiment Tracking¶

Script	Description
`logging_inference/multiple_loggers.py`	TensorBoard + CSV logging
`logging_inference/mlflow_tracking.py`	MLflow experiment tracking
`logging_inference/detailed_evaluation.py`	Confusion matrix and per-class metrics

CLI¶

Config	Description
`cli/classification.yaml`	CLI config for image classification
`cli/detection.yaml`	CLI config for object detection
`cli/segmentation.yaml`	CLI config for semantic segmentation

Deployment & Export¶

Script	Description
`deployment/export_to_onnx.py`	Export models to ONNX for cross-platform deployment
`deployment/export_to_torchscript.py`	Export models to TorchScript for production
`deployment/deploy_torchscript_cpp.py`	C++ and mobile deployment

Inference & Utilities¶

Script	Description
`logging_inference/inference.py`	Model inference and batch prediction
`logging_inference/inference_without_metrics.py`	Inference without metrics
`logging_inference/multilabel_inference.py`	Multi-label inference and per-label probabilities
`logging_inference/detection_inference.py`	Object detection inference
`logging_inference/segmentation_inference.py`	Segmentation inference
`logging_inference/backbone_discovery.py`	Explore timm backbones

Getting Started¶

To run any example:

# Clone the repository
git clone https://github.com/theja-vanka/AutoTimm.git
cd AutoTimm

# Install AutoTimm with all dependencies
pip install -e ".[all]"

# Run an example
python examples/getting_started/classify_cifar10.py

Example Categories Summary¶

Getting Started (3 examples)¶

Start your journey with CIFAR-10 classification and custom datasets.

Computer Vision Tasks (8 examples)¶

Classification - Basic to advanced classification
Object Detection - FCOS, YOLOX, RT-DETR, transformers (6 examples)
Segmentation - Semantic and instance segmentation (2 examples)

HuggingFace Hub (12 examples)¶

Basic Integration - 8 task-specific examples
Advanced Techniques - Interpretation, transfer learning, ensemble, deployment (4 examples)

Data & Training (14 examples)¶

Data & Augmentation - Balanced sampling, augmentation, multi-label, CSV loading (9 examples)
Training & Optimization - Auto-tuning, multi-GPU, HPO (5 examples)

CLI (3 configs)¶

Train from the command line with YAML configuration files.

Understanding & Deployment (13 examples)¶

Model Interpretation - GradCAM, attention, interactive viz (6 examples)
Logging & Tracking - TensorBoard, MLflow, evaluation (3 examples)
Inference & Utilities - Model inference, multi-label inference, and backbone discovery (4 examples)

Total: 50+ runnable examples and configs