How IntelinAir built the highest quality datasets with SuperAnnotate
IntelinAir uses the power of aerial imagery analytics, computer vision, deep learning, and mobile technology to deliver real-time, data-driven decision support to farmers, which helps them manage their operations more effectively.
IntelinAir was building pixel-accurate annotations for their aerial imagery annotation projects. They were searching for a platform that could deliver higher-quality annotations in less time.
SuperAnnotate’s platform, for richer tooling and more accurate annotations
IntelinAir has a wide range of annotation projects. Everything from parcel segmentation, crop detection, and plant/tree/flower counting. Many of these tasks have complex, pixel-accurate annotation requirements. IntelinAir built an in-house tool and explored other platforms previously, but was looking for a way to dramatically boost annotation speed and quality. That’s why they turned to SuperAnnotate.
Higher-quality annotations, increased efficiency, and the ability to annotate a project that was “not doable” before
IntelinAir used SuperAnnotate to create annotations that were more accurate than they ever could have accomplished with the previous tooling.
“One example of how we used SuperAnnotate was to create individual segmentation masks for a corn-kernel counting task,” Director of Machine Learning Jennifer Hobbs said. “The paintbrush functionality made this much easier (and more accurate) than we could have ever possibly accomplished in other tooling. I can’t imagine achieving this level of precision without SuperAnnotate’s platform.”
The greatest difference between SuperAnnotate and other tooling they have used in the past is in the quality of the annotation. “On past projects using different tools, even when we would try to get annotators to slow down and define smoother boundaries using more points, if the tooling wasn’t intuitive to use, the boundaries were still pretty coarse,” Hobbs said.
In addition, SuperAnnotate’s project management and workflow features helped improve the efficiency of IntelinAir’s projects. “I was able to upload the data, assign images to the (in-house) annotators, and then QA the data quickly,” Jennifer mentioned. “After we were done we were able to easily convert it to COCO format to release. The ability to define the hierarchical semantic nature of the classes is also very powerful.”
SuperAnnotate’s UI also made navigating through images quick and easy. “I particularly like the features around sorting, filtering, seeing the images along the bottom of the screen, hopping into an image from the main page, and seeing the analytics,” Jennifer noted.
With SuperAnnotate, IntelinAir is able to build pixel-accurate training data at quality levels and speeds unachievable with their previous tooling.
- SuperAnnotate’s feature-rich toolset helped IntelinAir create higher-quality annotations.
- With SuperAnnotate, IntelinAir was able to build significantly more accurate training data than was previously possible.
- The robust workflow features of SuperAnnotate led to more efficient project management.
Originally published at https://blog.superannotate.com.
Read more by SuperAnnotate:
Speed up image labeling using transfer learning (no code required)
How to automate and scale your image annotation process using transfer learning techniques.
Annotations for Aerial Imagery: Why Pixel Precision Will Be the New Norm
Guidelines on how to speed up pixel-accurate annotations with novel approaches.