AutoTimm - Automated Deep Learning for Computer Vision

Automated Deep Learning for Computer Vision

Powered by timm and PyTorch Lightning

Train state-of-the-art vision models with 1000+ backbones in just a few lines of Python

1

Install

pip install autotimm

Installation Guide →

2

Quick Start

Train your first model in minutes

Quick Start Guide →

3

Explore

Browse examples and dive deeper

View Examples →

Key Features¶

4 Vision Tasks
Image classification, object detection, semantic segmentation, and instance segmentation

1000+ Backbones
Any timm model: CNNs (ResNet, EfficientNet, ConvNeXt) and Transformers (ViT, Swin, DeiT)

Flexible Architectures
Built-in FCOS detector, DeepLabV3+ segmentation, Mask R-CNN style instance segmentation

Advanced Losses
Focal, Dice, Tversky, Combined CE+Dice, GIoU for bbox regression

Configurable Metrics
Use torchmetrics or custom metrics with full control

Multiple Loggers
TensorBoard, MLflow, W&B, CSV - use them all simultaneously

Auto-Tuning
Automatic learning rate and batch size finding

Enhanced Logging
Track learning rate, gradient norms, confusion matrices and more

Flexible Transforms
Torchvision (PIL) or albumentations (OpenCV) with bbox and mask support

Quick Example¶

import autotimm as at  # recommended alias
from autotimm import (
    AutoTrainer, ImageClassifier, ImageDataModule,
    LoggerConfig, MetricConfig,
)

# Data
data = ImageDataModule(
    data_dir="./data",
    dataset_name="CIFAR10",
    image_size=224,
    batch_size=64,
)

# Metrics (explicit configuration required)
metrics = [
    MetricConfig(
        name="accuracy",
        backend="torchmetrics",
        metric_class="Accuracy",
        params={"task": "multiclass"},
        stages=["train", "val", "test"],
        prog_bar=True,
    ),
]

# Model
model = ImageClassifier(
    backbone="resnet18",
    num_classes=10,
    metrics=metrics,
    lr=1e-3,
)

# Trainer with logging
trainer = AutoTrainer(
    max_epochs=10,
    logger=[LoggerConfig(backend="tensorboard", params={"save_dir": "logs"})],
    checkpoint_monitor="val/accuracy",
)

trainer.fit(model, datamodule=data)
trainer.test(model, datamodule=data)

That's it! Train production-ready models in ~20 lines of code

Why Choose AutoTimm?¶

Feature	AutoTimm	Raw PyTorch	Lightning
1000+ backbones	Yes	Manual	Manual
Configurable metrics	Yes	Manual	Manual
Multi-logger support	Yes	Manual	Partial
Auto LR/batch finding	Yes	No	Yes
Lines of code	~20	~200+	~100

Start Now →