Zetane Viewer is a free, cross-platform tool designed for visualizing the architecture and internal workings of machine learning models and artificial neural networks. It targets ML engineers, researchers, and practitioners seeking to demystify the "black box" of AI by inspecting tensors, weights, and layer outputs in 3D.

How It Works

The viewer supports both proprietary .ZTN (generated from Keras/PyTorch) and standard .ONNX model formats. It renders computational graphs, allowing users to navigate through interconnected nodes (operators) and tensors. Users can load custom inputs (.npy, .npz, .png, .jpg, etc.) to observe model inference in real-time and inspect individual layer components like feature maps, weights, and biases through a dedicated tensor view bar.

Quick Start & Requirements

• Install: Downloadable binaries for Windows, Linux, and Mac.
• Requirements: No specific hardware or software dependencies are listed beyond the OS.
• Resources: Links to download binaries, a tutorial video, and instructions for loading models are provided.

Highlighted Details

• Supports visualization of model architecture, intermediate tensors, weights, and biases.
• Allows loading custom input data in various formats (.npy, .npz, .png, .jpg, etc.).
• Offers multiple tensor visualization styles, including 3D and 2D projections with value-based coloring.
• Includes pre-generated ZTN and ONNX models for common architectures (Alexnet, Resnet, YoloV3, Unet, etc.) and tasks.

Maintenance & Community

No specific information on contributors, community channels (Discord/Slack), or roadmap is present in the README.

Licensing & Compatibility

The tool is described as "free," but a specific open-source license is not mentioned. Compatibility for commercial use or linking with closed-source projects is not detailed.

Limitations & Caveats

The README does not specify the open-source license, making commercial use and derivative works unclear. Information regarding community support, active maintenance, or a public roadmap is absent.