In 2022, we highlighted a few important computer vision trends, like edge computing, LiDAR technology, and explainable artificial intelligence (AI). This year, we polled thousands of artificial intelligence and computer vision experts and asked for their opinion.

This year promises to be even more exciting than last year! With the rise of generative technology and the expanded use of facial recognition, for example, 2023 is shaping up to bring even more advancements in the field of computer vision.

In this article, we explore the results and take a peek at what the top computer vision trends are for 2023:

  • Generative AI
  • Data-centric AI
  • Merged reality
  • Facial recognition
  • More accurate 3D models
Pie chart with the top computer vision trends in 2023

1. Generative AI

Generative artificial intelligence has the ability to create new and original content, instead of just analyzing or acting on existing data. ChatGPT is a well-known tool for text-generated information – DALL-E is commonly used for generating realistic photos and art from text prompts.

Stable Diffusion, a deep learning, text-to-image model, was released in 2022, helping multiple AI image generators emerge as competition. A famous image generator app, Lensa AI, uses artificial intelligence to turn photos into customized portraits.

Lensa AI utilizes images from users alongside the Stable Diffusion neural network to create high-quality digital portraits – and it can even mimic styles and artists. By using the deep learning model, Lensa AI can benefit from the ‘inpainting’ and ‘outpainting’ features. It can then ‘inpaint’ images with new variations, having a clear understanding of users’ facial features, ethnic background, and more.

Perhaps unsurprisingly if you’ve heard of generative AI’s capabilities, 76% of over 2,000 respondents also underlined this technology as the top trend for computer vision in 2023.

💡
Keep your eyes peeled, we’ve got a generative AI report in the works. 👀

2. Data-centric AI

Artificial intelligence is made of systems that are both model and data-centric. The latter is based on improving or changing datasets to increase a model’s performance. Model-centric artificial intelligence consists of an algorithm being updated while containing a fixed amount and type of data.

Fixed models have recently become more commonly used but it involves spending energy and time finding the best architecture models. Model-centric approaches have been popular in recent years but criticized for their limitation to consumer platforms.

When looking at a data-centric approach, you should keep in mind that it needs to be programmatic. This will help to cope with the large training data volume, with a programmatic process for both iterating and labeling data being essential.

Data-centric artificial intelligence is:

  • Cheaper
  • Faster to develop
  • Highly accurate

3. Merged reality

Augmented reality offers an interactive experience, combining computer-generated content with the real world. It can be used through smartphones, for example, and it helps to enhance both the real and the virtual world.

The popular game app, Pokémon GO, had the option of turning on the augmented reality feature for a more immersive experience when capturing Pokémon – and if nothing else, it showed us that people like to use augmented reality to interact with the world around them, especially when having fun.

Merged reality is very similar to augmented reality, as it doesn’t separate you from your surroundings – what it can do, is read what’s around you and add digital content. Like virtual reality, you need a headset to experience it.

Combining augmented reality with computer vision opens the way to very exciting developments. Simultaneous localization and mapping (SLAM) offer augmented reality systems with geometric positioning. This allows for the creation of 3D maps of environments through a camera’s location and position.

4. Facial recognition

Facial recognition scans and detects a person’s face from a database, matching it with either a video clip frame or a digital image. It pinpoints facial features in images and compares them with other images in a database with an artificial intelligence algorithm.

In healthcare, for example, we can expect to see this being used more in 2023. As facial recognition can automatically scan patients’ faces and extract all of their insurance and medical information, it can speed up medical professionals in their day-to-day procedures.

This technology is also helpful to diagnose medical disorders that would otherwise have difficult-to-detect symptoms. The Yellow Brick Road Project is working toward accelerating advancements for patients; FDNA developed Face2Gene. Helping medical professionals use a face analysis tool to help diagnose patients.


5. Bonus: More accurate 3D models

The creation of 3D models can often be extremely challenging, as it needs mechanical measurements and manual alignment of partial 3D views. With computer vision and artificial intelligence algorithms, you can take various stereo-pair images of a specific scene and automatically generate a geometrically accurate and photo-realistic digital 3D model.

Computer vision is able to produce these 3D models from image data and analyze the scene that’s projected into one or more images. This technology can help solve challenges like detecting the extent of distortion levels and faults, distinguishing distortions or flaws from color abnormalities, and even passing a pass/fail judgment depending on the capacity or volume.

Key takeaways

  • Generative AI: Generative artificial intelligence, capable of creating new, original content, has been identified by a majority of surveyed experts as a leading computer vision trend in 2023. Applications such as Lensa AI are using generative models like Stable Diffusion to create customized digital portraits with a deep understanding of users' features and backgrounds.
  • Data-centric AI: In contrast to model-centric AI, which emphasizes adjusting algorithms with a fixed dataset, data-centric AI focuses on modifying datasets for better model performance. The data-centric approach has been highlighted as being cheaper, quicker to develop, and highly accurate, making it a key trend in computer vision.
  • Merged Reality: Merged reality, a combination of the real world with computer-generated content, offers immersive experiences. With the potential for various applications, like games or spatial mapping (SLAM), this technology is an exciting trend for 2023.
  • Facial Recognition: This technology uses artificial intelligence algorithms to scan, detect, and match facial features from a database, with significant potential applications in healthcare. From expediting patient data retrieval to assisting in diagnosing medical disorders, facial recognition is expected to have a growing impact in the coming year.
  • More accurate 3D models: Computer vision and artificial intelligence are greatly enhancing the creation of accurate 3D models, with potential applications in detecting distortion levels, flaws, color abnormalities, and conducting capacity assessments. This trend of improving 3D modeling accuracy with computer vision is also gaining traction for 2023.