Browse » Benchmark


Roboflow hosts the most popular computer and machine vision benchmarking and transfer learning datasets. Datasets in this category include Microsoft COCO, Pascal VOC, MNIST, and more.


THE MNIST DATABASE of handwritten digits


  • Yann LeCun, Courant Institute, NYU
  • Corinna Cortes, Google Labs, New York
  • Christopher J.C. Burges, Microsoft Research, Redmond

Dataset Obtained From:

All images were sized 28x28 in the original dataset

The MNIST database of handwritten digits, available from this page, has a training set of 60,000 examples, and a test set of 10,000 examples. It is a subset of a larger set available from NIST. The digits have been size-normalized and centered in a fixed-size image.

It is a good database for people who want to try learning techniques and pattern recognition methods on real-world data while spending minimal efforts on preprocessing and formatting.

Version 1 (original-images_trainSetSplitBy80_20):

  • Original, raw images, with the train set split to provide 80% of its images to the training set and 20% of its images to the validation set
  • Trained from Roboflow Classification Model's ImageNet training checkpoint

Version 2 (original-images_ModifiedClasses_trainSetSplitBy80_20):

  • Original, raw images, with the train set split to provide 80% of its images to the training set and 20% of its images to the validation set
  • Modify Classes, a Roboflow preprocessing feature, was employed to change class names from 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 to one, two, three, four, five, six, seven, eight, nine
  • Trained from the Roboflow Classification Model's ImageNet training checkpoint

Version 3 (original-images_Original-MNIST-Splits):

  • Original images, with the original splits for MNIST: train (86% of images - 60,000 images) set and test (14% of images - 10,000 images) set only.
  • This version was not trained


                              title={MNIST handwritten digit database},
                              author={LeCun, Yann and Cortes, Corinna and Burges, CJ},
                              journal={ATT Labs [Online]. Available:},

Self Driving Car

  • Roboflow
  • obstacles Dataset
  • 13280 images


The original Udacity Self Driving Car Dataset is missing labels for thousands of pedestrians, bikers, cars, and traffic lights. This will result in poor model performance. When used in the context of self driving cars, this could even lead to human fatalities.

We re-labeled the dataset to correct errors and omissions. We have provided convenient downloads in many formats including VOC XML, COCO JSON, Tensorflow Object Detection TFRecords, and more.

Some examples of labels missing from the original dataset:
Examples of Missing Labels


The dataset contains 97,942 labels across 11 classes and 15,000 images. There are 1,720 null examples (images with no labels).

All images are 1920x1200 (download size ~3.1 GB). We have also provided a version downsampled to 512x512 (download size ~580 MB) that is suitable for most common machine learning models (including YOLO v3, Mask R-CNN, SSD, and mobilenet).

Annotations have been hand-checked for accuracy by Roboflow.

Class Balance

Annotation Distribution:
Annotation Heatmap

Use Cases

Udacity is building an open source self driving car! You might also try using this dataset to do person-detection and tracking.

Using this Dataset

Our updates to the dataset are released under the MIT License (the same license as the original annotations and images).

Note: the dataset contains many duplicated bounding boxes for the same subject which we have not corrected. You will probably want to filter them by taking the IOU for classes that are 100% overlapping or it could affect your model performance (expecially in stoplight detection which seems to suffer from an especially severe case of duplicated bounding boxes).

About Roboflow

Roboflow makes managing, preprocessing, augmenting, and versioning datasets for computer vision seamless.

Developers reduce 50% of their boilerplate code when using Roboflow's workflow, save training time, and increase model reproducibility.

Roboflow Wordmark

Pascal VOC 2012

Pascal VOC 2012 is common benchmark for object detection. It contains common objects that one might find in images on the web.

Image example

Note: the test set is witheld, as is common with benchmark datasets.

You can think of it sort of like a baby COCO.

Microsoft COCO

This is the full 2017 COCO object detection dataset (train and valid), which is a subset of the most recent 2020 COCO object detection dataset.

COCO is a large-scale object detection, segmentation, and captioning dataset of many object types easily recognizable by a 4-year-old. The data is initially collected and published by Microsoft. The original source of the data is here and the paper introducing the COCO dataset is here.