How lutzroeder/netron works
Netron lets a user open a machine-learning model and inspect its structure in a graph-oriented viewer. The documented format support spans many model ecosystems, so related readers are represented as format-family components rather than as a catalog of every adapter. The excerpts show desktop app actions reaching views and the view depending on graphing and shared model primitives. Parser-to-view dispatch is not shown in the sampled code, so those links are marked unverified; the README establishes the supported formats, not their internal routing.
Main entry points: view.js Model viewer · pytorch.js PyTorch reader · app.js Desktop app · tf.js TensorFlow readers
Components 20
User experience
- Desktop app
source/app.js - Model viewer
source/view.js
Model formats
- ONNX reader
source/onnx.js - PyTorch reader
source/pytorch.js - TensorFlow readers
source/tf.js - TensorFlow Lite reader
source/tflite.js - Core ML reader
source/coreml.js - Keras reader
source/keras.js - Other format readers
source
Model runtime
- Graph layout
source/grapher.js - Model primitives
source/base.js - Dagre layout engine
source/dagre.js
Data decoding
- Protobuf decoder
source/protobuf.js - FlatBuffers decoder
source/flatbuffers.js - HDF5 decoder
source/hdf5.js - ZIP decoder
source/zip.js - JSON decoder
source/json.js - XML decoder
source/xml.js - Text decoder
source/text.js
Other components
- Model user
Connections 20
- Model user → Desktop app: opens model
source/app.js - Desktop app → Model viewer: opens view
source/app.js - Model viewer → Graph layout: uses graphing
source/view.js - Model viewer → Model primitives: uses primitives
source/view.js - Graph layout → Dagre layout engine: uses layout
source/grapher.js - Model viewer → ONNX reader: dispatches to
- Model viewer → PyTorch reader: dispatches to
- Model viewer → TensorFlow readers: dispatches to
- Model viewer → TensorFlow Lite reader: dispatches to
- Model viewer → Core ML reader: dispatches to
- Model viewer → Keras reader: dispatches to
- Model viewer → Other format readers: dispatches to
source/view.js - PyTorch reader → Protobuf decoder: uses decoder
source/pytorch.js - PyTorch reader → FlatBuffers decoder: uses decoder
source/pytorch.js - PyTorch reader → ZIP decoder: uses archive
source/pytorch.js - PyTorch reader → Model primitives: uses primitives
source/pytorch.js - TensorFlow readers → Protobuf decoder: decodes messages
- HDF5 decoder → ZIP decoder: reads archive
source/hdf5.js - JSON decoder → Text decoder: uses decoder
source/json.js - XML decoder → Text decoder: uses decoder
source/xml.js