The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Back to feedPost
Deepfake Detection System
Problem: Deepfake accessibility creates real risks—misinformation, identity fraud, digital trust erosion. I built a production-grade detector: fast, accurate, explainable, CPU-deployable.
Approach: Transfer learning with ResNet-50 (25.5M params, pre-trained ImageNet-1K V2). Rather than training from scratch on limited data, I leveraged rich hierarchical features that transfer exceptionally well to detecting artifact patterns (blending boundaries, color mismatches, compression artifacts).
Key Decisions:
Face-Centric Preprocessing: OpenCV DNN crops faces to 224×224, concentrating attention on artifact-rich regions (eyes, mouth, jawlines).
Custom Classification Head: Bottleneck design (2048→512→1) with Focal Loss, label smoothing, progressive unfreezing—prevents overfitting on ~4,000 Ciplab images.
Explainable AI: Custom Grad-CAM generates attention heatmaps showing exactly where the model detected artifacts.
Real-World Robustness: Trained on JPEG compression, Gaussian noise, blur augmentations to handle compressed video feeds.
Results:
AUC-ROC: 0.9424 | Accuracy: 87.25% | Precision/Recall: 87.27%/84.81%
Inference: <100ms per face on CPU
Live on Hugging Face Spaces (no GPU)
Tech: PyTorch, OpenCV, Albumentations, Gradio. Images, videos, live webcam—one codebase.
Demonstrates: End-to-end AI engineering with deliberate trade-offs (efficiency over raw accuracy), training discipline (Focal Loss, Cosine Annealing), and production-first thinking (CPU compatibility, interpretability, real-world robustness).
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Trending
Claude
Claude has entered the design space. How are you using it?
Contra University
Learn from expert creatives how to earn more using next-gen AI tools.
Brand Design
The best brand designers are on Contra. Scroll to see what's trending in brand design. What are you building?
creativeaiflow
Creative AI workflows are evolving. What tools do you use, and what are their strengths and weaknesses?
freelancerlife
Freelancer life is wins, pivots, and everything in between. What’s yours right now?
Related posts
An instant, live-action meme generator. 🖼️🎻 Built this interactive AI Lens that runs smoothly until the user triggers it with a facial expression—instantly freezing the frame and repainting the image into a moody, 16th-century oil painting complete with a giant gold frame.
Transforming users into ancient, eroding archaeology! 🗿🌿 This advanced AI transformation deeply analyzes facial features to map a realistic cracked marble texture over the user, dynamically growing green moss and deep cracks directly within the crevices of their expressions.
Hype, hype and hype... It's the first page of the notebook that I write in tells me.
I was building automated workflows around Signalwatch, and it got me thinking about all the hype.
Hype around just one piece of the puzzle, LLMs. They just predict the next word while generating... No magic.The better they are trained, the better the output. Super helpful in a lot of contexts...
Tool calling. Helpful, but if LangChain and RAG + LLMs, for example, are not included where appropriate, ta ta, bye bye.
Automation. Someone bringing me a glass of water, for example, isn't something to be automated or waste resources or tokens on...
This brings me back to structural understanding, especially python + machine learning basics where you train a model based on a small piece of dataset you download from hugging face, for example... a lot becomes clear here
