About
CNN Visual Guide is an interactive explainer for how vision models work. 3Blue1Brown-style animations teach a concept; real trained models — from a 99% MNIST classifier up to a 76% top-1 ImageNet ResNet-50 — provide live evidence that the concept is real and that a specific architectural choice solves it.
The model progression
A "CNN fundamentals → limitations → residual fix → full scale" narrative, each step a real trained checkpoint.
grid_viewSimple CNN · MNIST99.2% Acc
layersDilated CNN · CIFAR-1088.4% Acc
account_treeResNet-18 · CIFAR-10076.6% Acc
rebase_editResNet-50 · ImageNet-1k76.01% top-1
hubDINOv3 ViT-L · Weapons · 9-class97.2% orig · 94.9% clean
view_in_arConvNeXt-V2-L · Weapons · 9-class96.5% orig · 96.9% clean
Dataset & model credits
- Datasets: MNIST, CIFAR-10, CIFAR-100, ImageNet-1k (respective licenses).
- Checkpoints trained from scratch in the ERA-4 course sessions.
- Built with Next.js, FastAPI, and PyTorch.
Image credits
External images are used under their Creative Commons licenses. Diagrams and data graphs not listed here were drawn for this guide.
- image2D convolution animation — Michael Plotke · CC BY-SA 3.0
- imageResidual block diagram — LunarLullaby (Wikimedia Commons) · CC BY-SA 4.0
- imageMNIST sample digits — Josef Steppan (Wikimedia Commons) · CC BY-SA 4.0
Contact
No accounts, no tracking. Uploaded images are processed in memory for inference and never stored.
© 2026 Moinuddin Hasan
← Back home