Akridata Named a Vendor to Watch in the IDC MarketScape for Worldwide Data Labeling Software Learn More
Finding Edge Cases
The Issue
Issue Statemtent
One of the challenging situations is when you have an example of a rare scenario and wanted to build a dataset to train your models to better handle such scenarios. However, finding more examples of such rare scenarios is easy said than done.
What exactly is the issue?
- Let’s say you’re a data scientist with a data set of 1,000,000 images that were collected by test vehicles.
- Out of all the images collected, you want to search and identify the 100 frames that represent a construction zone. One way to do this would be to downsample the dataset (100:1) to get to around 10,000 images and inspect for construction zone images. But sifting through 10,000 images is tedious and no trivial task.
- Unless another co-worker has already gone through the images to label each one with a Construction Zone, there is no way to search for the Construction Zone images without manually combing through all of the visual data. That’s a lot of time and effort to find a few select images.
The Akridata Solution
Here’s how Akridata Data Explorer helps to solve this issue.
Import your dataset
Select the dataset you want to work with
Let’s suppose you want to use our Patch Search feature to find images that have a traffic light.
You will see that the search results have faithfully captured the neighbouring frames in the video sequence that have the traffic light.
Explore your dataset
You can perform a variety of actions to understand what your dataset entails. You can get a summary view, look at images in each cluster groups, perform different choices of sampling or change number of clusters, etc.
Creating Your Novelty Training Set
Select the dataset you want to work with
Select the images matching the criteria you wish to train your model with.
In this scenario, since we are looking to train or improve autonomous cars’s ability to detect construction zones, we need to identify images containing construction, such as a barricade on the side of the road.
Intuitively provide postive and negative examples
Once you identify an image matching your desired criteria, you can select the “thumbs up” to give the model the positive reinforcement to return images closely related to the desired frame.
In tandem, you may also select “thumbs down” on non-relevant images to tell the model that you do not want images matching these criteria (such as an image with no visible construction zone or images taken at night).
Reviewing and Refining Search Results
Fine tune the search criteria
All of the images in the initial search results may not necessarily match the criteria “Construction Zone” as we did not explicitly state we are looking for such features.
By continuing this process of providing positive and negative reinforcement, the Data Explorer will provide higher accuracy results.
Saving the results
Add to or create a named Resultset
Once you create a resultset, you can add more images from new searches and remove unwanted images
Publish results
Publishing the resultset allows you to share with your team members in your organization and accessible for other downstream activities