Ultimate guide to image labeling for Recycling AI

Introduction to image labeling for recycling AI

In order to promote a circular economy and to combat pollution and climate change, many companies and governments today are working to promote recycling and to make it more efficient using AI. As one of the less attractive types of jobs, known as 4D: dull, dirty, dangerous, and difficult, manual waste processing and sorting is likely to be the first job to be fully replaced by robots. However, there is a long way to go, since 91% of plastic globally is not currently recycled. At the same time, 1 in every 5 items tossed for recycling are not suitable for the bin they’re thrown in.

To ensure that AI applications in this highly demanding field are reliable, it’s imperative to make sure selected datasets are as useful, representative, and diverse as possible. Recycling AI projects often deal with complex taxonomies which require the labelers to undergo a specialized training and to build up an expertise in distinguishing the different materials according to the chosen taxonomy. Our expertise with annotation for recycling purposes makes us, at Humans in the loop, an excellent choice to support you in your projects!

Challenges and best practices

Based on our experience annotating data for recycling AI applications, we’d like to share below some of the most recurring challenges one encounters in this type of work, and our best practices and recommendations on how to ensure the success of your AI project:

Challenge	Best Practice
Waste recycling solutions need to be localized for each country and even region where they are being applied. The wide variety of materials and brands around the world means that a model trained in one place will not be able to generalize to packages in another place.	In order to support our clients with building solutions which can be applied in different geographies, we build dedicated teams which learn the specificities of trash in each location and are able to distinguish between different products and brands. This expert knowledge becomes invaluable in order to build high-quality datasets which reflect local brands and packaging.
No matter how well the model performs, there will always be unknown or unidentifiable items in the sorting pipeline. These will end up miscategorized or put in a pile of “unidentifiable objects” which reduces the overall efficiency of the recycling process.	In order to handle unknown edge cases, Humans in the Loop offers a new solution for real-time feedback to AI systems. In its essence, every time an alert is triggered due to low model certainty, a human operator is called in order to determine the right classification of the object. Despite not being feasible at very large scales, this can be done for key unidentifiable objects in real time or post factum and the outputs from the human can be used to retrain the model and improve their accuracy.
The streams of materials submitted for recycling are very varied and may contain severe deformations and dirt. In addition, they may be occluded and cluttered. Finally, it is likely that new types of objects appear overtime, and there is a continuous need for new ground truth.	Challenging examples which are out of the usual distribution of the training data can be detected using active learning. This data is then channeled to human annotators for annotation so that they can generate new ground truth. This is a process which can be automated on a weekly or monthly basis so that outliers are consistently labelled and sent back for retraining the AI models so that they can be up to date.

This is why it’s really important to start with the right training data and to adopt a human-in-the-loop approach in order to be able to sift through large quantities of data and continuously improve and validate the recycling AI models you are using!

Types of annotation for Recycling AI

Below we are showcasing several different use cases of annotation for recycling and trash detection purposes. Many of these fall in the category of “AI for Good” and they receive the highest priority in our work, according to our Ethical AI policy.

Waste sorting for recycling plants

An increasing number of recycling facilities are adopting robots, picking arms, and smart conveyor belts in order to process the stream of incoming objects more efficiently. For such cases, we offer bounding box, polygonal, and instance segmentation of the objects.

Smart waste containers

AI systems can be applied earlier in the supply chain at the time of disposal, with smart waste containers which automatically categorize the incoming trash. For such use cases, bounding boxes with tags are the most appropriate approach.

Food waste monitoring in trash bins

Restaurants, cafes, supermarkets, and individual households can do a lot in order to reduce their food waste and recycle properly. With smart trash bins, waste can be monitored on a daily basis, offering insights and analysis on how to be more efficient.

Litter detection on streets

In a smart city setup, it will become increasingly commonplace to be able to detect litter and clean it up using automation. For such purposes, we offer precise bounding box or polygonal annotation with multi-level taxonomy in order to get a granular classification.

Trash detection on shores and in the water

The contamination of our shores, rivers, and seas is one of the major problems which we are facing today, and drone imagery can give us a comprehensive idea of the most contaminated locations. With polygonal or instance segmentation, and complex taxonomies for classifying the trash, this information can be processed at scale.

Blogs

Best Annotation Tools for 2021

Found your humans – now all you need is the right tool for the job? Here’s our review of the most popular tools for 2021!

How to use a human-in-the-loop for Recycling AI

In order to train and maintain a recycling AI solution, it’s a best practice to use a human-in-the-loop on a continuous basis in order to ensure high-quality training data and frequent iterations of model testing and improvement. Here are some of the ways in which humans can be plugged into the entire MLOps cycle:

Dataset collection: our humans in the loop on the ground can collect videos and images of different types of trash and litter in bins or in urban and natural environments, with a wide variety of materials, settings, angles, camera types, and locations
Ground truth annotation: in order to train your initial models, we offer full dataset annotation from scratch in batches: bounding box annotation on images and video, as well as more complex polygonal and semantic or instance segmentation tasks.
Output validation with active learning: once you’ve trained an initial model, we can use it in order to pre-annotate a large part of the dataset, which will both increase the speed of the annotators and the impact of their work, by setting up an active learning workflow and prioritizing instances where your model is least certain
Adversarial example collection: once you’ve trained an initial model, we can expand your core dataset with additional difficult and challenging edge cases, such as unusual types of materials, extreme deformations, or challenging videos or camera angles, depending on the failure modes of your model during testing
Real-time edge case handling: once you have a model in deployment, our humans-in-the-loop are available 24/7 to handle potential edge cases that appear in real time or close to real time, using a simple API request and sending the correct response in seconds in order to ensure a second layer of verification for your model’s most critical responses

Wondering who is annotating your data?

When you are hiring a company to help you with your annotation needs, you frequently never meet the workers who are labeling your data. We want to change this and present to you the inspiring stories of our annotators!

Does this sound like something you’d like to try out? Get in touch with us and we’d be happy to schedule a free trial so as to explore how we can best help you with your recycling annotation needs!