Draw.io Neural Network Model Library

A collection of 159 ready-to-use neural network architecture templates for draw.io, designed for academic papers, technical reports, and presentations. All diagrams follow IEEE Transactions color conventions with a consistent, professional style.

How to Use

Quickest: open the library in draw.io with one click, nothing to download. For offline use, grab nn_model_library.xml or the full zip from the latest release.

Open nn_model_library.xml in draw.io (File > Open Library) to load all 159 templates into the scratchpad. Drag any template onto your canvas and customize as needed.

Each category is also available as a smaller library in libraries/, and nn_components.xml holds single blocks, operators, arrows and a color legend for drawing your own architectures.

Alternatively, individual .drawio files are available in exports/drawio/ for standalone use.

Preview

CNN Architectures

LeNet-5	ResNet Block	MobileNet Block	Inception Module

Transformer Architectures

Vision Transformer (ViT)	Transformer Encoder	Swin Transformer Block

Sequence Models

Seq2Seq + Attention	BERT Block	GPT Block

Detection & Segmentation

U-Net	Faster R-CNN

Full Template List (159 Models)

CNNs: LeNet-5, AlexNet, VGG Block, ResNet Block, ResNet Bottleneck, DenseNet Block, MobileNet Block, EfficientNet MBConv, Inception Module, SE Block, CBAM, ConvNeXt Block, ResNeXt Block, Depthwise Separable Conv, ShuffleNet V2 Unit, RepVGG Block, Ghost Module, HRNet Stage

Transformers: Transformer Encoder, Transformer Decoder, Full Transformer, ViT, Swin Transformer Block, Multi-Head Attention Detail, MLP-Mixer, Scaled Dot-Product Attention, Cross-Attention Block, Pre-LN vs Post-LN, Attention Masks, Sparse Attention Patterns, Linear Attention, Swin Transformer, DeiT, PVT Stage, Perceiver IO, MaxViT Block, Patch Embedding

NLP Models: BERT Block, GPT Block, Seq2Seq, Seq2Seq + Attention, LLaMA Decoder Block, Grouped-Query Attention, Multi-Head Latent Attention, MoE Layer, SwiGLU FFN, Rotary Position Embedding, Decoder-Only LLM, T5 Encoder-Decoder

RNNs: Stacked LSTM, Stacked GRU, Bi-LSTM, LSTM Cell, GRU Cell

Sequence Models: TCN Residual Block, WaveNet, Conformer Block, Mamba Block, Selective SSM

Detection & Segmentation: Faster R-CNN, FPN, YOLO Head, U-Net, FCN Decoder, ASPP Module, DETR, Mask R-CNN, RetinaNet, SSD, YOLOv8, C2f Block, PANet, CenterNet, DeepLabv3+, PSPNet, SAM, Mask2Former, SegFormer, Attention U-Net, UNet++

Generative Models: GAN, DCGAN Generator, DCGAN Discriminator, Conditional GAN, VAE, Autoencoder, VQ-VAE, VQGAN, StyleGAN Generator, CycleGAN, Pix2Pix, RealNVP Coupling Layer, PixelCNN

Diffusion Models: DDPM Process, Diffusion U-Net, Time-Conditioned ResBlock, Latent Diffusion, DiT Block, ControlNet, Classifier-Free Guidance, Flow Matching

Multimodal: CLIP, BLIP-2, Q-Former, LLaVA, Flamingo, Early vs Late Fusion, Whisper

Self-Supervised: SimCLR, MoCo, BYOL, SimSiam, MAE, DINO, Barlow Twins, Triplet Network

Graph Neural Networks: GCN Layer, GraphSAGE, GIN, Message Passing, Graph Classification Pipeline, Graph Transformer, ST-GCN

Reinforcement Learning: Agent-Environment Loop, DQN, Dueling DQN, Actor-Critic, PPO, AlphaZero, Dreamer

3D Vision: PointNet, PointNet++ Set Abstraction, DGCNN EdgeConv, NeRF, 3D Gaussian Splatting, VoxNet, Multi-View CNN

Scientific ML: PINN, Neural ODE, KAN Layer, Fourier Neural Operator, DeepONet, Capsule Network

Recommender Systems: Wide & Deep, DeepFM, DCN, DLRM, Two-Tower, DIN

Training & Systems: LoRA, KV Cache Decoding, Speculative Decoding, RLHF Pipeline, DPO, RAG Pipeline, ReAct Agent Loop, Knowledge Distillation

Other: GAT Layer, Siamese Network

Color Scheme

All templates use a standardized IEEE Transactions academic color palette.

Component	Fill	Stroke
Attention	#E1D5E7	#9673A6
Convolution	#DAE8FC	#6C8EBF
Deconvolution	#DCEEF8	#56A5C9
RNN (LSTM/GRU)	#D4EDDA	#28A745
Pooling	#D5E8D4	#82B366
Normalization	#F5F5F5	#999999
FC / MLP	#FFE6CC	#D79B00
Input	#F8CECC	#B85450
Output	#FFF2CC	#D6B656
Operators	#FFFFFF	#666666
State Space / Memory	#D5F0EE	#4A9C94
Gating / Routing	#F9DCEB	#B5739D

License

This library is provided under the MIT License. Feel free to use these templates in your papers, presentations, and projects.