Fashion Use Case

Our experienced and professional image annotaters draw bounding boxes around the objects in question so that our data is of high quality, ready for machine learning model training. As you can see in the example image here, some objects are chosen while the rest is left out. This process is done repeatedly on thousands of images using our distributed work force. And then, our project managers check over every image to ensure the highest quality.

Semantic Segmentation

Used by the most accurate clothing recognition models, semantic segmentation labels each pixel in a complex scene with class and attributes.

Instance Segmentation

When more bounding boxes are having a diminishing return on clothing recognition performance, models can benefit from using highly accurate masks.

Instance Segmentation

Each bounding box encompasses the clothing item an is labeled with its category and all its various attributes.


How the data was collected

Our first step in creating a dataset for fashion is to collect data. We scour the planet for tens of thousands of original images, content created for us from micro workers around the world. They are paid in Ether for their hard work, which is diligently scrutinized by our experienced project managers. Once we have quality image data then we can analyze the images in our image annotation platform, by our annotater team, also distributed around the world.

We are currently looking for Data Scientists, Integration Engineers, DLT Engineers, Android Developers, Full Stack Javascript Developers, and Machine Learning Engineers.

See The Dataset In Action

Here we have a high quality dataset - a large group of diverse images in the wild. Unlike photos on the web, which are often marketing images with photosohopped backgrounds, our images are truly real-world images. They include the complexities of real world lighting and backgrounds and so on, which makes this data all the stronger.

Model Training and Evaluation

At this point we have trained the model ourselves on various platforms, such as Google AutoML and Amazon, and have evaluated the results in our own platform. As you can see here we can demonstrate the precision, recall, confidence threshold, and more aspects of the model. Also below you can see a confusion matrix, which shows the accuracy in recognizing specific objects that we have trained the model to recognize. But don’t worry, we customize our efforts to tailor to your level of needs, so whether or not you have machine learning engineers, we can help you.

Let's Get In Touch

We are available to speak with you regarding your needs for any of our products or services. Let us know about your project and let’s get in touch!

Contact Us