I'm currently working on my final year project, and my supervisor has mentioned that I need to enhance it beyond simply using APIs. I completely understand, but figuring out how to improve what I've built is proving to be a challenge. Many of my peers have created impressive projects that combine existing technologies and tools, like autonomous cars, but I'm feeling a bit stuck.
My project focuses on a DeepFake API designed for social media platforms. The API's main function is to analyze photos for deepfaked faces, blurring them out if detected, and returning the modified image. I also implemented a feature allowing users to upload their own face photos, which are stored as vector embeddings. When an image is sent for verification, if a matching or similar embedding is found, it also blurs that user's face.
I've managed to implement these features, but I'm uncertain about where to go next in terms of improvement.
Here's what I'm currently using:
- InsightFace for face detection and recognition
- A deepfake predictor built through transfer learning using EfficientNetV2, which has achieved a 70% accuracy rate.
I've considered a few directions for improvement:
1. Enhancing the deepfake predictor model, though I'm aware industry standards might outperform it significantly.
2. Improving face recognition, possibly through one-shot face creation to handle different angles, but I've found that similar solutions already exist.
I'm just unsure of how to truly enhance existing technologies or demonstrate deeper development in my project.
3 Answers
You're doing an impressive project at your stage! The key might be in how you present it rather than just the technical components. If you can show thorough testing, like having confidence scores for each detected face or implementing a review aspect for false positives, that will definitely impress your supervisor. They’re likely pushing you to think bigger; don't hesitate to explore that!
Your project sounds solid as it is! A good way to show more depth might be to add an evaluation pipeline to analyze precision and recall with different datasets. You could also conduct adversarial tests or compare various deepfake models while explaining the performance differences of yours. Supervisors usually appreciate seeing a comprehensive workflow, so consider adding detection processes along with confidence scores, logging, or a review pipeline for flagged images.
I’m really impressed by what you’re doing! Just remember, everyone starts somewhere. Keep learning and iterating on your ideas. You'll definitely find ways to enhance your work. You got this!
Related Questions
How To: Running Codex CLI on Windows with Azure OpenAI
Set Wordpress Featured Image Using Javascript
How To Fix PHP Random Being The Same
Why no WebP Support with Wordpress
Replace Wordpress Cron With Linux Cron
Customize Yoast Canonical URL Programmatically