tensorspace-team/tensorspace

Features

• Neural Network 3D Renderers - Renders the full structural layout of neural networks using 3D graphics to illustrate data flow.
• WebGL Rendering - Uses the WebGL API to render neural network layers and tensors as interactive 3D objects.
• Activation Visualizations - Renders visual representations of internal layer activations to analyze feature maps based on input images.
• Deep Learning Model Renderers - Converts model definitions into animated 3D components to visualize data flow using WebGL.
• Dense Layer Rendering - Renders fully connected neural network layers with configurable node counts and interactive animations.
• Pre-trained Weight Loading - Imports external model files and maps them to visual layers via a standardized preprocessing step.
• 1D Convolutional Rendering - Renders 1D convolutional layers in 3D with configurable filters, kernel sizes, and strides.
• 2D Convolutional Rendering - Renders 2D convolutional layers as 3D objects to illustrate image feature extraction.
• ML Model Format Converters - Converts pre-trained models from external frameworks into formats compatible with the 3D renderer.
• Model Architecture Visualizations - Renders the structural layout and layer connectivity of Keras models in a 3D space.
• Model Inspectors - Provides a visual interface for running inference and analyzing intermediate activations within ML models.
• Model Topology Visualizations - Maps the layers and connections of a machine learning model into an interactive visual representation.
• Neural Layer Merging Visualization - Provides visual representations of mathematical operations that combine multiple neural network layers.
• Neural Network Visualizations - Generates graphical 3D representations of neural network architectures, layers, and connectivity.
• Animated Forward Pass Visualizations - Provides interactive 3D visualizations that animate the step-by-step data flow during a neural network forward pass.
• Pre-trained Model Importers - Allows the import of existing model architectures from external libraries to create immediate 3D representations.
• TensorFlow Model Inference - Executes TensorFlow model predictions in the browser to visualize real-time intermediate activations.
• Interpretability - Displays internal tensor data and feature maps to help users understand how models generate predictions.
• Architecture Definitions - Enables the definition of network structures through sequential arrangements of convolutional, pooling, and dense layers.
• Keras - Transforms pre-trained Keras models into a compatible format for high-fidelity 3D visualization.
• Visualization Schema Conversions - Transforms external Keras and TensorFlow model architectures into an internal format for 3D spatial visualization.
• 2D Activation Rendering - Renders two-dimensional matrices of activation function outputs with toggleable visibility and animated transitions.
• Layer-Specific Rendering Logic - Provides dedicated 3D drawing mechanisms tailored to different tensor shapes and neural layer types.
• 1D Activation Rendering - Renders one-dimensional activation functions using configurable shapes and pagination for long layers.
• 1D Input Rendering - Renders one-dimensional data input layers with configurable shapes and adjustable animation speeds.
• 1D Output Rendering - Renders one-dimensional output layers with configurable node counts, labels, and pagination.
• Neural Network 3D Visualization Frameworks - Provides a browser-based 3D framework for rendering interactive representations of neural network architectures and tensor data.
• Intermediate Data Visualizers - Displays internal activations and tensor states of hidden layers to visualize how outputs are generated.
• Feature Map Aggregators - Collapses multiple data channels into simplified representations using max or average values for easier visualization.
• Layer Dimension Inspection - Tracks and retrieves input and output tensor shapes for each neural network layer.
• Convolution Layers - Renders transposed convolution layers as interactive 3D components with customizable aesthetics.
• Depthwise Convolutions - Renders depthwise spatial convolution layers where filters are applied per channel in a 3D space.
• Intermediate Node Extraction - Retrieves data from internal model nodes to render them as 3D objects for detailed analysis.
• Model Prediction Visualizations - Generates visual explanations and animations that map model predictions back to the flow of information.
• Object Detection Feature Map Renderers - Renders the prediction grids and probability regions specifically used in YOLO object detection models.
• 1D Pooling Rendering - Renders the structural operation of 1D max or average pooling using configurable windows and strides.
• 2D Pooling Rendering - Renders the structural transformations of 2D max or average pooling using window sizes and padding.
• Reshaping Visualizations - Visualizes the transformation of tensor data across 1D, 2D, or 3D spatial layouts.
• Activation Inspectors - Runs inference in the browser to analyze and visualize intermediate activations and data flow.
• Neural Network Topology Layouts - Provides settings to define how model feature maps are arranged and aggregated for 3D rendering.
• Feature Map View Toggles - Provides interactive toggles to switch between aggregated and expanded views of neural network feature maps.
• Architecture Analysis - Converts pre-trained Keras architectures into spatial representations to inspect layer dimensions and network topology.
• 2D Tensor Rendering - Renders 2D data matrices as abstract layers with custom shapes and appearance.
• Coordinate-Based Topology Mapping - Calculates 3D spatial coordinates and connection paths between neural network layers for architectural visualization.
• Detection Result Rendering - Draws bounding boxes and labels on visualization layers to represent object detection results.
• Image Input Visualizations - Renders greyscale images as the initial input layer of a neural network pipeline in 3D.
• Neural Layer Animations - Manages visual state transitions and animations for neural network architectural components.
• Neural Layer Visualizations - Provides the ability to define abstract 3D layers with user-specified shapes to visualize external data.
• Inference Monitors - Monitors and displays real-time layer metrics and operational statistics during model inference.
• Layer Aggregation Toggles - Provides UI controls to switch between detailed expanded views and simplified aggregated states of 3D feature maps.
• Tensor Visualization Pagination - Splits oversized one-dimensional tensor layers into multiple visual segments to maintain rendering performance.
• Architecture Prototype Visualizers - Visualizes the 3D structure of model prototypes without the need for trained weights.
• Deep Learning Frameworks - 3D visualization framework for neural networks in browsers.
• Model Visualization - Renders interactive 3D visualizations of neural networks in browsers.

Star history