lutzroeder/netron

Feature visualization

Open

#687 opened on Feb 24, 2021

View on GitHub
 (4 comments) (3 reactions) (0 assignees)JavaScript (32,743 stars) (3,101 forks)batch import
featurehelp wanted

Description

add feature visualization to netron

you select/upload a image from the file system. When you sleect a layer, netron displays the features that maximise the activation of the selected layer

based on the following research paper: https://arxiv.org/pdf/1311.2901.pdf

there was a similar repo (https://github.com/keplr-io/quiver), but it was written for tensorflow 1 and it fails to construct valid maps of the models constructed with modern tools

this might be network-specific, so maybe implement slowly?

Contributor guide

Tech stack
javascripttensorflow
Domain
machine learningfrontend
Issue type
feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
4
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
over 1 week
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
stale
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
mostly clear
Prerequisites
Basic knowledge of neural networksFamiliarity with Netron codebaseJavaScript programming
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
35
Research direction
Study the referenced paper (arXiv:1311.2901) for visualization techniques. Examine the quiver repository for inspiration, noting it was written for TensorFlow 1. Review Netron's source code to understand how models are loaded and layers are selected. Implement a feature to upload an image and compute activations for a selected layer, then display the maximizing features. Network specific handling may be needed, so consider starting with a simple model.