Mountain Dew Commercial Computer Vision Project

Drop an image or


Mountain Dew is running a $1,000,000 counting contest. Computer Vision can help you win.


Watch our video explaining how to use this dataset.

Mountain Dew

During Super Bowl LV, Mountain Dew sponsored an ad that encourages viewers to count all unique occurrences of Mountain Dew bottles. You can watch the full ad here. The first person to tweet the exactly correct count at Mountain Dew is eligible to win $1 million (see rules here).

Counting things is a perfect place for where computer vision can help.

We uploaded the Mountain Dew video to Roboflow, created three images per each second of the commercial (91 images from ~30 seconds of commercial), and annotated all bottles we could see. This dataset is the result.

We trained a model to recognize the Mountain Dew bottles, and then ran the original commercial back through this model. This helps identify Mountain Dew bottles that the human eye may have missed when completing counts.

Image example

Getting Started

Click "Fork" in the upper right hand corner or download the raw annotations in your desired format.

Note that while the images are property of PepsiCo, we are using them here as fair-use for educational purposes and have released the annotations under a Creative Commons license.

About Roboflow

Roboflow enables teams to use computer vision.
Our end-to-end platform enables developers to collect, organize, annotate, train, deploy, and improve their computer vision models -- all without needing to hire a new ML engineering team.

Roboflow Wordmark

Cite this Project

If you use this dataset in a research paper, please cite it using the following BibTeX:

@misc{ mountain-dew-commercial_dataset,
    title = { Mountain Dew Commercial Dataset },
    type = { Open Source Dataset },
    author = { Roboflow },
    howpublished = { \url{ } },
    url = { },
    journal = { Roboflow Universe },
    publisher = { Roboflow },
    year = { 2021 },
    month = { feb },
    note = { visited on 2022-10-04 },




Joseph Nelson

Last Updated

2 years ago

Project Type

Object Detection






CC BY 4.0