LIVECell dataset

This document contains instructions of how to access the data associated with the submitted manuscript “LIVECell - A large-scale dataset for label-free live cell segmentation” by Edlund et. al. 2021.


Light microscopy is a cheap, accessible, non-invasive modality that when combined with well-established protocols of two-dimensional cell culture facilitates high-throughput quantitative imaging to study biological phenomena. Accurate segmentation of individual cells enables exploration of complex biological questions, but this requires sophisticated imaging processing pipelines due to the low contrast and high object density. Deep learning-based methods are considered state-of-the-art for most computer vision problems but require vast amounts of annotated data, for which there is no suitable resource available in the field of label-free cellular imaging. To address this gap we present LIVECell, a high-quality, manually annotated and expert-validated dataset that is the largest of its kind to date, consisting of over 1.6 million cells from a diverse set of cell morphologies and culture densities. To further demonstrate its utility, we provide convolutional neural network-based models trained and evaluated on LIVECell.

How to access LIVECell

All images in LIVECell are available following this link (requires 1.3 GB). Annotations for the different experiments are linked below.

LIVECell-wide train and evaluate

Annotation set URL
Training set link
Validation set link
Test set link

Single cell-type experiments

Cell Type Training set Validation set Test set
A172 link link link
BT474 link link link
BV-2 link link link
Huh7 link link link
MCF7 link link link
SH-SHY5Y link link link
SkBr3 link link link
SK-OV-3 link link link

Dataset size experiments

Split URL
2 % link
4 % link
5 % link
25 % link
50 % link

Comparison to fluorescence-based object counts

The images and corresponding json-file with object count per image is available together with the raw fluorescent images the counts is based on.

Cell Type Images Counts Fluorescent images
A549 link link link
A172 link link link

Download all of LIVECell

The LIVECell-dataset and trained models is stored in an Amazon Web Services (AWS) S3-bucket. It is easiest to download the dataset if you have an AWS IAM-user using the AWS-CLI in the folder you would like to download the dataset to by simply:

aws s3 sync s3://livecell-dataset .

If you do not have an AWS IAM-user, the procedure is a little bit more involved. We can use curl to make an HTTP-request to get the S3 XML-response and save to files.xml:

curl -H "GET /?list-type=2 HTTP/1.1" \
     -H "Host:" \
     -H "Date: 20161025T124500Z" \
     -H "Content-Type: text/plain" > files.xml

We then get the urls from files using grep:

grep -oPm1 "(?<=<Key>)[^<]+" files.xml | sed -e 's/^/http:\/\/\//' > urls.txt

Then download the files you like using wget.

File structure

The top-level structure of the files is arranged like:

    ├── LIVECell_dataset_2021  
    |       ├── annotations/
    |       ├── models/
    |       ├── nuclear_count_benchmark/	
    |       └──  
    └── LICENSE


The images of the LIVECell-dataset are stored in /livecell-dataset/LIVECell_dataset_2021/ along with their annotations in /livecell-dataset/LIVECell_dataset_2021/annotations/.

Within are the training/validation-set and test-set images are completely separate to facilitate fair comparison between studies. The images require 1.3 GB disk space unzipped and are arranged like:

    ├── livecell_test_images
    |       └── <Cell Type>
    |               └── <Cell Type>_Phase_<Well>_<Location>_<Timestamp>_<Crop>.tif
    └── livecell_train_val_images
            └── <Cell Type>

Where <Cell Type> is each of the eight cell-types in LIVECell (A172, BT474, BV2, Huh7, MCF7, SHSY5Y, SkBr3, SKOV3). Wells <Well> are the location in the 96-well plate used to culture cells, <Location> indicates location in the well where the image was acquired, <Timestamp> the time passed since the beginning of the experiment to image acquisition and <Crop> index of the crop of the original larger image. An example image name is A172_Phase_C7_1_02d16h00m_2.tif, which is an image of A172-cells, grown in well C7 where the image is acquired in position 1 two days and 16 hours after experiment start (crop position 2).


The annotations of LIVECell are prepared for all tasks along with the training/validation/test splits used for all experiments in the paper. The annotations require 2.1 GB of disk space and are arranged like:

    ├── LIVECell
    |       └── livecell_coco_<train/val/test>.json
    ├── LIVECell_single_cells
    |       └── <Cell Type>
    |               └── <train/val/test>.json
    └── LIVECell_dataset_size_split
            └── <Split>_train<Percentage>percent.json

All annotations are in Microsoft COCO Object Detection-format, and can for instance be parsed by the Python package pycocotools.


ALL models trained and evaluated for tasks associated with LIVECell are made available for wider use. The models are trained using detectron2, Facebook’s framework for object detection and instance segmentation. The models require 15 GB of disk space and are arranged like:

   └── Anchor_<free/based>
            ├── ALL/
            |    ├── <Model>.pth
            |    └── config.yaml       
            ├── <Cell Type>/
            |    ├── <Model>.pth
            |    └── config.yaml
            ├── <Base-Config>.yaml

Where each <Model>.pth is a binary file containing the model weights, <Base-Config>.yaml the basic Detectron2-config each model training’s config.yaml inherits and a README explaining how to set up and use the model.


The images and fluorescence-based object counts are stored as the label-free images in a zip-archive and the corresponding counts in a json as below:

    ├── A172_counts.json
    ├── A549_counts.json 

The json files are on the following format:

    "<filename>": "<count>"

Where <filename> points to one of the images in the zip-archive, and <count> refers to the object count according fluorescent nuclear labels.


All images, annotations and models associated with LIVECell are published under Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) license.