🔌 CNN-to-Linear Shape Mismatch Visualizer
Watch the image shrink, unfold, and plug in — glowing green only when the numbers truly match.
📋 How To Use This Lab
Adjust the
Image Size
,
Kernel/Stride/Padding
, and
MaxPool
sliders — every shape updates live down the pipe.
Watch the block physically
shrink
at each stage, then
unfold
into a flat line at Flatten.
Type a guess into
in_features
— if it's wrong, the plug shows red and disconnected, just like PyTorch's real error.
Press
✓ Auto-Calculate
to snap in_features to the exact correct value and watch the plug glow green.
Flattened = 3136
✓ Connected
Input Image
3 × 28 × 28
Conv2d
k=3, s=1, p=1
16 × 28 × 28
MaxPool2d
pool=2×2
16 × 14 × 14
Flatten
16 × 14 × 14 = 3136
nn.Linear Input Socket
🎛️ Architecture Parameters
Input Image Size:
28
×
28
Conv Kernel Size:
3
Conv Stride:
1
Conv Padding:
1
MaxPool Size
2 × 2
3 × 3
nn.Linear(in_features =
)
✓ Auto-Calculate
🔌 PLUGGY
Type a number into in_features — I'll show you exactly whether it plugs in cleanly.
🧑🏫 PROF. TORCH
in_features must equal out_channels × H_final × W_final — nothing more, nothing less.