How fast does it take to label an image with 3,000 objects?
It's all about the numbers
Problem
An aerial imagery company needed to annotate a dataset of aerial images that included up to ~3,000 objects per image. The goal was to detect all of the objects, mark each of them with an accurate bounding box and to classify each object from a list of options.
They initially tried to manage it internally only to find out that it takes many hours and is tedious, exhausting and frustrating. When they tried to outsource it through BPOs, they found it expensive, time consuming and the quality wasn’t that great either.
Solution
By using Tasq.ai, our customer was able to receive the images loaded with objects faster and at better quality than they anticipated. Due to Tasq.ai’s patent-pending microtasking workflow technology, our platform deconstructed each image into hundreds of micro-tasks to be labeled by our Tasqers. We gamified the tasks to create interest and avoid ‘boredom’, limiting each Tasqer only to a few minutes of consecutive work in order to avoid burnout and exhaustion. Each image was annotated simultaneously by hundreds of Tasqers, with multiple judgements and qualifications resulting in high quality results faster than any other solution. This revolutionized our client’s data labeling process.
The Workflow
The Workflow built by our Customer Success team was as follows:
- Break down every image into a grid consisting of more than 500 images with overlays to avoid losing any object in between.
- Build a task for spotting, identifying and marking every object with multiple dynamic judgements for ensuring confidence scores.
- Automatically labeling of bounding boxes around every marked object using ML for high confidence.
- Human validation and correction of each bounding box to ensure max quality results.
- Human classification of every object per the list of prior classes.
- Aggregation of all results and confidence scores per annotation to one single structured image.
Result
High quality annotated dataset at unmatched speed, improving models, enabling our client to reach production faster and reach their business goals faster.
