Skip to content

Model Gallery

This page showcases visualizations of classic deep learning models generated using torchview.

All images are in SVG vector format. You can click on any image or zoom in your browser to view all details.

πŸ–ΌοΈ Computer Vision

ResNet-50

ResNet-50

Residual Network: A classic CNN architecture that solves the training challenges of deep networks through skip connections. The Bottleneck structure is clearly shown in the visualization.

Inception V3

Inception V3

Inception Architecture: Known for its multi-scale branches (Inception Modules), this architecture is both complex and efficient.

Vision Transformer (ViT)

ViT

ViT: Applies the Transformer architecture directly to image patches, completely eliminating convolutional operations.

FCN (Fully Convolutional Network)

FCN

Fully Convolutional Network: A classic model for semantic segmentation, built on a ResNet-50 backbone.

πŸ“ Natural Language Processing (NLP)

Transformer Encoder (BERT-style)

Transformer Encoder

Transformer Encoder: Contains Multi-Head Attention and Feed Forward Network, serving as the foundation for models like BERT.

Seq2Seq with Attention

Seq2Seq

Sequence-to-Sequence with Attention: A classic machine translation architecture. The visualization shows the Encoder (left) and Decoder (right) with the Attention computation process in between (horizontal layout).

πŸ›’ Recommender Systems

Wide & Deep

Wide & Deep

Wide & Deep: A classic recommendation architecture that combines linear models (Wide, for memorization) with deep neural networks (Deep, for generalization).

🎨 Generative Models

VAE (Variational Autoencoder)

VAE

Variational Autoencoder: The visualization shows the complete generative pipeline: Encoder β†’ Reparameterization (sampling) β†’ Decoder.

⭐ Special Architectures

WaveNet

WaveNet

WaveNet: A generative model for speech synthesis. The visualization clearly shows the stacked residual blocks based on dilated convolutions (horizontal layout).

GCN (Graph Convolutional Network)

GCN

Graph Convolutional Network: A neural network for processing graph-structured data.

How to Contribute

If you'd like to add new models to the gallery, please ensure: 1. The model architecture is representative. 2. Use torchview to generate SVG format for clarity. 3. Submit a PR to our repository.