![NYU Researchers Develop AI System for Accurate Age Modification in Images NYU Researchers Develop AI System for Accurate Age Modification in Images](https://aiheadliner.com/wp-content/uploads/2023/08/Screenshot-2023-08-31-at-1.20.14-AM.png)
The Significance of AI in Estimating and Modifying Age
AI systems are increasingly being used to accurately estimate and modify the ages of individuals through image analysis. To build models that can handle aging variations, a large amount of data and high-quality longitudinal datasets are required. These datasets consist of images of a large number of individuals collected over several years.
Challenges and Solutions
While numerous AI models have been designed for age estimation and modification, many face challenges in effectively manipulating age while preserving the individual’s facial identity. One common challenge is assembling a large set of training data that shows individual people over many years.
However, researchers at NYU Tandon School of Engineering have developed a new AI technique to change a person’s apparent age in images while preserving their unique biometric identity. They trained the model using a small set of images of each individual and a separate collection of images with captions indicating the person’s age category. These captioned pictures help the model understand the relationship between images and age. The trained model can then simulate either aging or de-aging scenarios by specifying a desired target age through a text prompt.
The Training Process and Evaluation
The researchers utilized a pre-trained latent diffusion model, a small set of training face images of an individual, and a small auxiliary set of image-caption pairs to fine-tune the model using appropriate loss functions. They also added and removed random variations in the images. The researchers used the “DreamBooth” technique, which involves a gradual and controlled transformation process facilitated by a fusion of neural network components, to manipulate the human facial images.
To evaluate the accuracy of the model, the researchers compared it to alternative age-modification techniques. They asked 26 volunteers to associate the generated images with actual photographs of the same individuals. The comparison also included the use of ArcFace, a facial recognition algorithm. The results showed that the researchers’ method outperformed other techniques, reducing the frequency of incorrect rejections by up to 44%.
Findings and Conclusion
The researchers found that when the training dataset includes images from the middle-aged category, the generated images effectively represent a diverse range of age groups. However, if the training set primarily consists of images from the elderly category, the model struggles to generate pictures of individuals in the opposite age extremes, such as the child category. The generated images also showed a better capability to transform the training images into older age groups for men compared to women. This difference may be due to the inclusion of makeup in the training images. However, variations in ethnicity or race did not noticeably affect the generated outputs.
In conclusion, AI techniques for estimating and modifying age in images have the potential to revolutionize various fields, including biometrics and entertainment. The development of robust models that can accurately manipulate age while preserving individual identity is crucial for advancing this technology further.
Read the full research paper here.
Join our 29k+ ML SubReddit, 40k+ Facebook Community, Discord Channel, and Email Newsletter for the latest AI research news, cool AI projects, and more.