NVIDIA Feature Map Explorer (FME) enables visualization of 4-dimensional image-based feature map data using a range of views, from low-level channel visualizations to detailed numerical information about each channel slice. It provides deep learning developers with intimate information about what the model is learning, where the model is failing to use resources efficiently, and what is changing as a model is learning during training to better process data handed to it.

Features

• Visualize Feature Maps at Multiple Levels: Generate a range of analysis data, including low-level channel visualizations and detailed numerical information about the full feature map tensor and each channel slice.
• Check for Opportunities to Improve Speed: Know when you can drop to lower precision numerical formats to improve speed.
• Understand Training Progressions: Easily compare feature maps from one epoch to another to understand training progressions.
• Compatible with Any Training Network: Works with feature map data stored in standard numpy .npy files available from any training network.
• Dive Into Each Channel: Visualize and examine each channel to see what has been learned or what is missing in a feature.
• Compare Feature Maps: Efficiently inspect the processing taking place across the layers of a deep learning model to guide changes to the model or training parameters.
• Get Better Inference Results: Understand the efficiency and potential deficiencies of a model to optimize training parameters and improve inference results.

Benefits

• Enhanced Model Insights: Provides detailed information about what the model is learning and where it is failing, helping developers optimize their models.
• Improved Speed: Identifies opportunities to drop to lower precision numerical formats, improving processing speed.
• Training Progressions: Facilitates easy comparison of feature maps across epochs to understand training progressions.
• Versatile Compatibility: Compatible with any training network, making it a flexible tool for deep learning developers.
• Channel-Level Analysis: Allows deep learning developers to dive into each channel to see detailed information about learned features.
• Optimized Inference: Helps developers understand model efficiency and deficiencies to optimize training parameters and improve inference results.