CS5720 - Week 4
Slide 77 of 80

Translation Invariance

What is Translation Invariance?

Translation invariance means that a CNN can recognize objects regardless of where they appear in the image. A cat is still a cat whether it's in the top-left or bottom-right corner.
How CNNs achieve this:

• Parameter sharing across spatial locations
• Same filters applied everywhere
• Pooling operations reduce spatial sensitivity
• Multiple layers build spatial robustness
💡 Key Insight
Translation invariance is crucial for real-world vision because objects can appear anywhere in an image, and we want consistent recognition.

Real-World Examples

  • 👤
    Face Detection
    Faces detected regardless of position in photo
  • 🚗
    Object Recognition
    Cars recognized anywhere in traffic images
  • 🔬
    Medical Imaging
    Tumors detected regardless of location in scan
  • 📝
    Text Recognition
    Letters recognized anywhere in document

Interactive Translation Invariance Demo

Position 1: Top-Left
Detection: ✓ Cat
Position 2: Center
Detection: ✓ Cat
Position 3: Bottom-Right
Detection: ✓ Cat
Notice: The CNN correctly identifies the object regardless of its position!
← Previous Next →
Prepared by Dr. Gorkem Kar