Logo

Getting Started

  • Main concepts
    • Self-supervised Learning
  • Installation
    • Supported Python versions
    • Installing Lightly
    • Dependencies
    • Next Steps
  • Command-line tool
    • Check the installation of lightly
    • Crop Images using Labels or Predictions
    • Training and Embedding in a Go – Magic
    • Train a model using the CLI
    • Create embeddings using the CLI
    • Download data using the CLI
    • Breakdown of lightly-magic
  • Self-supervised learning
    • How Lightly Works
      • Data and Transformations
      • Model, Loss and Training
      • Embeddings
    • Lightly in Three Lines
    • What’s next?

Advanced

  • Advanced Concepts in Self-Supervised Learning
    • Augmentations
    • Previewing Augmentations
    • Models
    • Losses
      • Memory Bank
    • Obtaining Good Embeddings
      • Monitoring Embedding Quality
    • Extracting specific Video Frames
  • Distributed Training
    • Distributed Training Benchmarks
      • Observations
  • Benchmarks
    • ImageNette
    • CIFAR10
    • Imagenet100
    • Next Steps

Tutorials

  • Python Package
    • Tutorial 1: Structure Your Input
      • Supported File Types
        • Images
        • Videos
      • Image Folder Datasets
        • Flat Directory Containing Images
        • Directory with Subdirectories Containing Images
      • Video Folder Datasets
      • Embedding Files
        • Advanced usage of Embeddings
      • Next Steps
    • Tutorial 2: Train MoCo on CIFAR-10
      • Imports
      • Configuration
      • Setup data augmentations and loaders
      • Create the MoCo Lightning Module
      • Create the Classifier Lightning Module
      • Train the MoCo model
      • Next Steps
    • Tutorial 3: Train SimCLR on Clothing
      • Imports
      • Configuration
      • Setup data augmentations and loaders
      • Create the SimCLR Model
      • Visualize Nearest Neighbors
      • Color Invariance
      • Next Steps
    • Tutorial 4: Train SimSiam on Satellite Images
      • Imports
      • Configuration
      • Setup data augmentations and loaders
      • Create the SimSiam model
      • Train SimSiam
      • Scatter Plot and Nearest Neighbors
      • Next Steps
    • Tutorial 5: Custom Augmentations
      • Imports
      • Configuration
      • Setup custom data augmentations
      • Setup dataset and dataloader
      • Create the MoCo model
      • Train MoCo with custom augmentations
      • Evaluate the results
    • Tutorial 6: Pre-train a Detectron2 Backbone with Lightly
      • Introduction
      • Prerequisites:
      • Imports
      • Configuration
      • Initialize the Detectron2 Model
      • Setup data augmentations and loaders
      • Self-supervised pre-training
      • Storing the checkpoint
      • Finetuning with Detectron2
      • Next Steps
  • Platform
    • Tutorial 1: Curate Pizza Images
      • What you will learn
      • Requirements
      • Upload the data
      • Filter the dataset using metadata
      • Download the curated dataset
      • Training a model using the curated data
    • Tutorial 2: Diversify the Sunflowers Dataset
      • What you will learn
      • Requirements
      • Create a Selection
      • Download selected samples
    • Tutorial 3: Active learning for classification
      • What you will learn
      • Define your dataset
      • Creation of the dataset on the Lightly Platform with embeddings
      • Active learning
    • Tutorial 4: Active Learning using Detectron2 on Comma10k
      • Requirements
      • Upload dataset to Lightly
      • Inference on unlabeled data
      • Create our Detectron2 model
      • Get Model Predictions
      • Query Samples for Labeling
      • Next Steps
    • Tutorial 5: Custom Metadata and Rebalancing
      • What you will learn
      • Requirements
      • Custom Metadata
      • Configuration
      • Rebalancing
    • Tutorial 6: Find false negatives of object detection
      • What you will learn
      • Requirements
      • Steps
    • Tutorial 7: Active Learning with Nvidia TLT
    • Tutorial 8: Integration with LabelStudio for Active Learning
    • Tutorial 9: Embedded COVID mask detection
    • Tutorial 10: Export to LabelStudio
      • What you will learn
      • Requirements
      • Launch LabelStudio
      • Export from Lightly in the LabelStudio format
      • Import the tasks into LabelStudio
      • Start labeling

Examples

  • Models
    • Barlow Twins
    • BYOL
    • DCL & DCLW
    • DINO
    • MAE
    • MSN
    • MoCo
    • NNCLR
    • PMSN
    • SimCLR
    • SimMIM
    • SimSiam
    • SMoG
    • SwaV
    • SwaV Queue
    • TiCo
    • VICReg
    • VICRegL

Python API

  • lightly
  • lightly.api
    • .api_workflow_client
  • lightly.cli
    • .lightly_cli
    • .train_cli
    • .embed_cli
    • .download_cli
    • .version_cli
    • .crop_cli
    • .config.config.yaml
      • Overwrites
      • Additional Arguments
      • Default Settings
  • .core
  • lightly.data
    • .collate
    • .dataset
  • lightly.loss
    • .barlow_twins_loss
    • .dcl_loss
    • .dino_loss
    • .hypersphere_loss
    • .memory_bank
    • .msn_loss
    • .negative_cosine_similarity
    • .ntx_ent_loss
    • .regularizer.co2
    • .pmsn_loss
    • .sym_neg_cos_sim_loss
    • .swav_loss
    • .tico_loss
    • .vicreg_loss
    • .vicregl_loss
  • lightly.models
    • .resnet
    • .zoo
    • .nn_memory_bank
    • .heads
  • lightly.transforms
    • .gaussian_blur
    • .rotation
    • .solarize
  • lightly.utils
    • .io
    • .embeddings_2d
    • .benchmarking
    • .debug
    • .dist
    • .reordering
    • .version_compare

On-Premise

  • Docker Archive
    • Setup
      • Analytics
      • Licensing
      • Download the Docker Image
      • Sanity Check
      • Update Lightly Docker
    • First Steps
      • Storage Access
      • Specify Relevant Files
      • Embedding a Dataset and Selecting from it
      • Train a Self-Supervised Model
      • Accessing Lightly Input Parameters
      • Selecting from Embeddings File
      • Manually Inspecting the Embeddings
      • Selecting from Video Files
      • Removing Exact Duplicates
      • Upload Sampled Dataset To Lightly Platform
      • Reporting
        • Live View of Docker Status
        • Docker Output
        • Evaluation of the Selection Proces
    • Advanced
      • Meta Information
        • Mask Samples
        • Use Pre-Selected Samples
        • Custom Weak Labels
      • Datapool
        • How It Works
        • Usage
      • Pretagging
        • How It Works
        • Usage
      • Add Predictions to a Datasource
        • Predictions Folder Structure
        • Prediction Tasks
        • Prediction Schema
        • Prediction Files
        • Prediction Files for Videos
        • Prediction Format
        • Prediction Singletons
        • Creating the predictions folder
      • Add Metadata to a Datasource
        • Metadata Folder Structure
        • Metadata Schema
        • Metadata Files
        • Metadata Format
        • Next Steps
      • Active Learning
        • Prerequisites
        • Selection
        • Active Learning with Custom Scores (not recommended as of March 2022)
      • Sequence Selection
        • How It Works
        • Usage
      • Object Level
        • Prerequisites
        • Predictions
        • Selection on Object Level
        • Lightly Pretagging
        • Padding
        • Object Crops Dataset
        • Analyzing the Crop Dataset
        • Multiple Object Level Runs
    • Integration
      • Using the Docker with a Cloud Bucket as Remote Datasource
        • Introduction
        • Advantages
        • Requirements
        • Download the Lightly Docker
        • Run the Lightly Docker with the datasource
        • View the progress of the Lightly Docker
        • Use your selected dataset
        • Process new data in your bucket using a datapool
        • Network traffic
      • Trigger a Docker Job from from the Platform or code
        • Introduction
        • Advantages
        • Download the Lightly Docker
        • Register the Lightly Docker as a Worker
        • Create a Dataset and Trigger a Job
        • View the progress of the Lightly Docker
        • Use your selected dataset
        • Process new data in your bucket using a datapool
      • Load data directly from S3 buckets using s3fs-fuse
        • What is s3fs-fuse?
        • Get an AWS Bucket and Credentials
        • Install s3fs-fuse
        • Configure S3 Credentials
        • Use S3 Storage with Lightly Docker
        • Use Caching
        • Common Issues
      • Data Pre-processing Pipeline on AWS with Dagster
        • Introduction
        • Dagster
        • Setting up the EC2 Instance
        • Setting up the S3 Bucket
        • Integration
    • Configuration
      • List of Parameters
      • Choosing the Right Parameters
      • Increase I/O Performance
    • Examples
      • Extract Diverse Video Frames
        • Using ffmpeg
        • Using Lightly Docker
      • ImageNet
      • Combining Cityscapes with Kitti
    • Known Issues and FAQ
      • Docker is slow when working with long videos
      • Docker Crashes when running with GPUs
      • Shared Memory Error when running Lightly Docker
      • Docker crashes because of too many open files
      • Permission denied for input created with sudo
      • Error when using S3 fuse and mounting to docker
      • Token printed to shared stdout or logs
    • Hardware recommendations
      • Finding the compute speed bottleneck
      • Updating the machine
lightly
  • Docker Archive
  • Integration
  • View page source

Integration

Here you learn how to integrate the Lightly Docker into data pre-processing pipelines.

  • Using the Docker with a Cloud Bucket as Remote Datasource
  • Trigger a Docker Job from from the Platform or code
  • Load data directly from S3 buckets using s3fs-fuse
  • Data Pre-processing Pipeline on AWS with Dagster
Next Previous

© Copyright 2020, Lightly AG

Built with Sphinx using a theme provided by Read the Docs.