Is a "Face Mask Detector" project too dated to put on a Computer Vision resume in late 2023?

All answers to this question.

I would definitely retire the mask detector. In 2023, "Action Recognition" or "Anomaly Detection" is much more relevant. For example, build a system that detects "unusual behavior" in a security camera feed—like someone falling or a fire starting. This requires understanding temporal data, not just static frames. Another great path is "Edge AI"—optimizing a YOLO model to run efficiently on a Raspberry Pi or a mobile phone. Companies love to see that you can make models fast and light, not just accurate on a high-end GPU.

   Answered 2023-09-07 by Melissa Bryant

Melissa, when you say "Temporal Data," do you mean I should be looking into LSTM-CNN architectures, or are Vision Transformers (ViTs) the way to go for these types of projects now?

   Answered 2023-09-09 by Steven Cook

• Steven, while LSTMs were the standard, Vision Transformers are definitely where the research is headed. However, for a beginner project, a 3D-CNN or a simple Frame-Difference method is often more stable and easier to explain in an interview. If you can implement a ViT and explain why it’s better for your specific dataset, you’ll look like an absolute pro. Just make sure you can explain the "Attention" mechanism clearly!

     Commented 2023-09-10 by Melissa Bryant

Consider a "Synthetic Data" project. Show how you used tools like Unity or Blender to create training data for a CV model. It’s a very hot skill right now.

   Answered 2023-09-12 by Jason Scott

• Exactly, Jason. Data scarcity is a huge problem in Computer Vision. Showing you can generate high-quality synthetic data to augment a small real-world set is a brilliant portfolio addition.

     Commented 2023-09-13 by Karen Walker