Competitive processes usually produce the best outcomes, and deep learning for computer vision is no different. Platform.ai Competitions are a new way to solve the toughest computer vision challenges with minimal upfront risk.
Competition sponsors define the challenge, supply a few public sample images as examples, and label a ground truth set of images using Platform.ai which serves as the benchmark of accuracy among competitive entries. We recruit data scientists and relevant domain experts to compete, and they use Platform.ai to build and submit entries for evaluation.
Platform.ai Competition offer a number of key benefits to sponsors over other model-building solutions:
Complete confidentiality for your corporate data: Competition entrants never get to see corporate image data, unless sponsors want them to. Confidential images are used to measure the accuracy of the submitted models, which is done without entrants ever seeing the images themselves.
More generalizable models: Because the competition entrants are aggregating large sets of training images as they build their model, and because we aggregate this training data, the models produce by Platform.ai Competitions are more generalizable to real-world conditions than model produced on in-house images alone.
Speed to Final Model: Because Platform.ai is a rapid development platform, each competitor can build, refine and submit many models in a short amount of time. Our typical competition is one month in length.
Low Risk Way to Build Models: Depending on the time and complexity, building a computer vision model can run hundreds of thousands of dollars in time and expenses before you even get to deployment. Platform.ai Competition fees start as low as $10,000, derisking the model creation process.Learn More
COMPETITIONS ARE THE IDEAL MODEL-BUILDING SOLUTION FOR:
Companies who want to build computer vision apps but don’t have enough training data of their own
Companies who want to build computer vision apps but can’t let outsiders see their corporate data due to confidentiality, HIPAA or any other reason
Companies that have hit an accuracy wall on a current computer vision model and want to increase their accuracy further
Companies whose training image data is curated and not reflective of the range of real-world images which the model will be used with
Companies that want to test the viability of a computer vision applications but want to limit their risk in initial development costs
Whether you’re a trained data scientist, an expert in the subject of the competition, or anyone else with data wrangling powers, you could earn money, prestige, and bragging rights by building a more accurate model.
To enter our competitions, here’s what you need to do:
1) Establish a platform.ai account
2) Enter the competition by (doing the following):
2.1) Go to the "/competitions" page by clicking on the "COMPETITIONS" button at top menu.
2.2) Choose which competition you would like to join in "Active Competitions" list and click on it.
2.3) At the competition description page click on the "JOIN COMPETITION" button:
2.4) Choose a project which you want to use in this competition or create a default project with a default competition collection:
If you decided to choose an existing project, make sure it has the same labels as the competition.
After you joined the competition, it will available in "My Competitions" section of the "/competitions" page.
Once you’re in the competition, start sourcing your own training data and use Platform.ai to label it and build initial models. To train a model on your currently labeled data, click on the button.
You may submit a model as an entry once every 2 hours. To submit the entry, click on the button. At this point, your model will be evaluated based on the accuracy of its predictions against a portion of the sponsor’s ground truth images, and your entry in the Competition leaderboard will be updated (or added if its your first submission).
We encourage all contestants to make active use of our community channels – brainstorming between Platform,.ai and contestants and among contestants can help improve performance for all.See competitions list
How quickly can you build my model?
Our standard Competitions are one month. Add a few weeks for setting up the competition and promoting to the entrants. Once built, the model is immediately deployable.
What do I get as a Competition sponsor?
The winning model is licensed exclusively to you. Depending of the level of sponsorship, this model can run on the Platform.ai service, or exported to any other platform or device.
Who competes in Platform.ai Competitions?
Platform.ai has some users that are data scientists, and other users that are passionate experts about certain domains, and both enter competitions that appeal to them. Since Platform.ai requires no coding capabilities, anyone that is up to the challenge could participate once they’ve registered.
My corporate images are confidential and can’t be shown outside of the company. Can I still take advantage of your competitions?
Yes. You never need to share your restricted image files with the entrants to the competition unless you want to. You’d work with your own files on Platform.ai to create the reference model for evaluation, but these images are never seen by the entrants.
Does the Platform.ai team work to improve the models produced by the Competitions?
We provide technology to improve model performance, such as out-of-domain image filtering, and can also provide additional augmentation work if desired. We spend a lot of team in each competition helping the entrants understand what’s driving and limiting their model performance, so that they can improve in their next submission.
Why are Platform.ai Competitions a better route for me to take than other competitive services?
Platform.ai is a much faster way to build image classification models than any other tool, so our Competitions move faster. Our entrants submit between 30-40 times in a month, compared to the 2-3 submissions per month you see on other services.
How is the winner selected?
The winning entry is the model submitted by a contestant which has the highest accuracy in classifying a sample set of images from the reference ground truth image set, which is labeled by the sponsor.