I created a pretty niche image viewer, similar to <code>tev</code>.
<a href="https://github.com/pkubik/paw-viewer" target="_blank" rel="noopener nofollow">https://github.com/pkubik/paw-viewer</a>
<img src="https://media.daily.dev/image/upload/s--jPjWZaVi--/f_auto/v1769813943/ugc/content_c471dccb-3a6a-41e4-818a-ae99940921a3?_a=BAMAMiiu0" alt="paw">
<a href="https://youtu.be/j7iVhv--PGg" target="_blank" rel="noopener nofollow">(YouTube)</a>
Recently, I’ve been working on image upscaling and denoising models.
Part of my workflow was to load outputs of several models in to <code>tev</code> (a niche image viewer for this kind of applications) and look for changes in quality, such as spatial or temporal artifacts.
While <code>tev</code> is a great viewer it was lacking certain quality-of-life features, such as directly loading <code>numpy</code> arrays, saving crop coordinates or more advanced Python interoperability (<code>tevclient</code> is very basic).
I implemented a <code>paw-viewer</code> to fill in the missing features and to create a solid base for hacking similar tools.
Features:
<ul>
<li>Loading and displaying images and videos in popular formats</li>
<li>Zoom with nearest neighbor interpolation</li>
<li>Video playback with precise control</li>
<li>Easy switching between multiple sources, e.g. your algorithm outputs and reference images</li>
<li>Selecting regions for extracting cropped data to clipboard or a file
<ul>
<li>This includes small JSON text with crop definition (for use in QA automation configs)</li>
</ul>
</li>
<li>Controlling exposure and gamma</li>
<li>Easy Python API for passing <code>numpy</code> arrays</li>
</ul>

Paweł Kubik

A specialized Python image viewer called paw-viewer was created for comparing outputs from ML models, particularly for image upscaling and denoising work. It improves upon tev by adding features like direct numpy array loading, crop coordinate saving, video playback controls, and enhanced Python interoperability. The tool supports multiple image formats, precise zoom with nearest neighbor interpolation, exposure/gamma controls, and easy switching between algorithm outputs and reference images.

Niche Python image viewer for comparing video sequences from ML models