Top Neural Networks Visualization Tools
As human beings, we are more comfortable looking at pictures to comprehend instead of trying to mentally visualize complex structures like deep neural networks. Also, it is more convenient to present visually and explain your new architecture to a wider audience which is not as tech “savy” as you are 🙂
Keeping this in mind, today, in this article, I am listing down top neural networks visualization tool which you can use to see how your architecture looks like visually. Without further ado, let us begin.
1. Netron
Project URL: https://github.com/lutzroeder/netron
Model type supports:
- ONNX (.onnx, .pb, .pbtxt)
- Keras (.h5, .keras)
- Core ML (.mlmodel)
- Caffe (.caffemodel, .prototxt)
- Caffe2 (predict_net.pb)
- Darknet (.cfg)
- MXNet (.model, -symbol.json)
- Barracuda (.nn)
- ncnn (.param)
- Tengine (.tmfile)
- TNN (.tnnproto)
- UFF (.uff)
- TensorFlow Lite (.tflite)
Experimental model supports:
- TorchScript (.pt, .pth)
- PyTorch (.pt, .pth)
- Torch (.t7)
- Arm NN (.armnn)
- BigDL (.bigdl, .model)
- Chainer (.npz, .h5)
- CNTK (.model, .cntk)
- Deeplearning4j (.zip)
- MediaPipe (.pbtxt)
- ML.NET (.zip)
- MNN (.mnn)
- PaddlePaddle (.zip, model)
- OpenVINO (.xml)
- scikit-learn (.pkl)
- TensorFlow.js (model.json, .pb)
- TensorFlow (.pb, .meta, .pbtxt, .ckpt, .index)
2. Neuroph
Project URL: http://neuroph.sourceforge.net/
Neuroph Studio Edition: https://github.com/neuroph/NeurophStudio
Neuroph, the Java-based framework, other than NN visualization, also supports class diagram visualization with a rich and well-documented set of APIs.
Supported neural network architectures:
- Adaline
- Perceptron
- Multi-Layer Perceptron with Backpropagation, Momentum on Resilient Propagation
- Hopfield network
- Bidirectional Associative Memory
- Kohonen network
- Hebbian network
- Maxnet
- Competitive network
- Instar
- Outstar
- RBF network
- Neuro-Fuzzy Reasoner
3. Neural Designer
URL: https://www.neuraldesigner.com/
Neural Designer is a commercial machine learning platform with rich interactive UI to design and visualisation of AI ML models.
4. Deep Learning Studio
URL: https://deepcognition.ai/
Project type: Commercial (Limited free version available)
DeepLearning Studio claims to be mostly coding free. That is, you would almost require zero coding to visualize and get your DNN model ready.
DeelLearning Studio’s simple drag & drop interface helps you design deep learning models with ease. Pre-trained models, as well as use built-in assistive features, simplify and accelerate the model development process. You can import model code and edit the model with the visual interface.
- deepcognition.ai
5. NVIDIA DIGITS
Project URL: https://developer.nvidia.com/digits
DIGITS is available as a free download to the members of the NVIDIA Developer Program.
The NVIDIA Deep Learning GPU Training System (DIGITS) puts the power of deep learning into the hands of engineers and data scientists. DIGITS can be used to rapidly train the highly accurate deep neural network (DNNs) for image classification, segmentation and object detection tasks. It has the following key features:
- Design, train and visualize deep neural networks for image classification, segmentation and object detection using Caffe, Torch and TensorFlow
- Download pre-trained models such as AlexNet, GoogLeNet, LeNet and UNET from the DIGITS Model Store
- Perform hyperparameter sweep of learning rate and batch size for improved model accuracy
- Schedule, monitor, and manage neural network training jobs, and analyze accuracy and loss in real time
- Import a wide variety of image formats and sources with DIGITS plug-in
- Scale training jobs across multiple GPUs automatically
6. ENNUI
Project URL: https://github.com/martinjm97/ENNUI
App URL: https://math.mit.edu/ennui/
ENNUI is an Elegant Neural Network User Interface that allows users to:
- Build a neural network architecture with a drag-and-drop interface.
- Train those networks on the browser.
- Visualize the training process.
- Export to Python.
Subscribe to my free newsletter to regularly receive bite-sized ML articles breaking complex concepts into small, simple consumable bites to become ML upskilled slowly but surely.
