If suits and ties are not your cup of tea, getting dressed for work might yield some interesting results from time to time. While sticking out for exceptional performance at work is a good thing, getting attention because you wore shorts on the day an important client was in town is not. So if you do not know the difference between appropriate business attire and the “I just came from the beach” look, you might need some assistance.
A pair of students at the National Institute of Technology Warangal has developed a solution that may help. They have created an artificial intelligence (AI)-powered device that uses computer vision to look you over. It then compares what you are wearing to a predefined dress code and determines if you are good to go, or if you need to try a little bit (or a lot) harder.
Samples from the training dataset (📷: L. Durgam et al.)
To find the best solution for this application, the team developed a custom convolutional neural network (CNN), and also retrained a MobileNetV2 model, to determine which option would be more accurate. The ultimate goal of the project was to design a real-time dress code detection system that could help schools and workplaces maintain standards of appropriate attire.
To build and train their models, the students used Edge Impulse, a machine learning development platform designed for edge devices. They created a custom image dataset containing examples of acceptable and unacceptable clothing and used transfer learning to adapt the pre-trained MobileNetV2 model to their specific needs. They also trained several CNN models with varying input image sizes ranging from 64×64 to 160×160 pixels.
Once trained, the models were deployed on an NVIDIA Jetson Nano, a compact, energy-efficient device capable of running AI models locally. This edge deployment allows the system to function in real time without needing to send data to the cloud, and that preserves user privacy while maintaining fast execution speeds.
The dress code detection system (📷: L. Durgam et al.)
During operation, the system captures images or video footage of individuals using a camera, and the onboard AI model analyzes the clothing. The model then classifies the outfit as either compliant or non-compliant with the predefined dress code. If a violation is detected, the system can generate an alert or report for administrators.
The team compared the performance of their CNN models and the retrained MobileNetV2 model in terms of accuracy, latency, storage use, and processing speed. The MobileNetV2 model achieved the highest test accuracy at 98.6%. However, it required more memory and had higher latency than simpler CNN models. If really fast detections are required, that would make the custom CNNs a better choice, but when it comes to dress code detection, waiting a second or two would generally be fine.
In combining AI, computer vision, and edge computing, the students have created a practical and scalable solution for dress code enforcement. If your personal standard is not quite so high, and you just want to make sure you are wearing pants for video calls, then Safe Meeting has got you covered.