CNN X-Ray Vision
cnn • vision • feature-maps
🩻 CNN X-Ray Vision
CNN picture ko “human eyes” jaise nahi dekhta. Ye chhote-chhote pattern robots se scan karta hai.
🎯 Quest
Draw a standing line | so 🧱 Vertical detector wins!
Score
0
Streak
0
Badge
—
🗣️ Narrator
Ready! Draw or pick a preset, then Start X-Ray 🚀
🧰 Input tools
Grid:
Paint:Draw karo ✍️
Brush:
Image → grayscale → downsample → grid (demo). Then Start X-Ray.
🖍️ Input
White pixel = “ink”. Heat glow (step 3) = “CNN looked here more”.
SAME padding ON ✅
Why feature map size sometimes smaller? Because stamp must fully fit. SAME padding adds a zero border so output stays same size.
🧠 Feature Maps
Green = kernel found its pattern ✅, Red = opposite pattern 🙃. Bigger = stronger.
Selected: 🧱 Vertical Edge
map size: 12×12
Main khada danda ( | ) dhoondhta hoon. Left vs Right compare karta hoon.
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
🧠 Hover Explanation (kid mode)
Feature-map ke kisi cell pe hover karo. Main tumhe EXACT bataunga:
✅ kaunsa 3×3 patch liya
✅ kernel ne kya multiply kiya
✅ sum karke number kaise bana
✅ kaunsa 3×3 patch liya
✅ kernel ne kya multiply kiya
✅ sum karke number kaise bana
🔴 Red vs 🟢 Green (super simple)
🟢 Green number: robot bola “YES! mujhe mera pattern mila!”
🔴 Red number: robot bola “ye toh ulta pattern hai” (opposite edge)
Bigger magnitude = stronger feeling.
🔴 Red number: robot bola “ye toh ulta pattern hai” (opposite edge)
Bigger magnitude = stronger feeling.