Skip to content

Instantly share code, notes, and snippets.

@blaz-r

blaz-r/GSOC2023.md Secret

Created August 22, 2023 16:19
Show Gist options
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save blaz-r/cfaa95ff33708635fefade1d12b7a8a7 to your computer and use it in GitHub Desktop.
Save blaz-r/cfaa95ff33708635fefade1d12b7a8a7 to your computer and use it in GitHub Desktop.
Download ZIP
Add ensembling methods for tiling to Anomalib (Google Summer of Code 2023)
Raw
GSOC2023.md

Add ensembling methods for tiling to Anomalib (Google Summer of Code 2023)

Contributor: Blaž Rolih (@blaz-r)

Mentors: Dick Ameln (@djdameln), Samet Akcay (@samet-akcay)

Organization: OpenVINO Toolkit (GSoC page, GitHub page)

GSoC Project page

GitHub repository: Anomalib

Pull request: Tiled Ensemble PR

Tiled Ensemble

Detecting defects in high-resolution images poses challenges, including limitations due to memory constraints. Downsampling leads to information loss and potentially missing small defects. A way to address this is the so-called tiled ensemble approach, where we split the image into smaller tiles and train separate models for each tile position. This way information is preserved, performance is improved, and memory constraints are managed.

In this GSoC project, tiled ensemble approach was implemented as part of Anomalib, a deep-learning anomaly detection library.

More information and insights from the entire journey can be found in the Anomalib discussions thread.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment