Keuper Labs

Papers needing PDFs

The website has a detail page for every listed publication. The following seven full PDFs could not be downloaded on 2026-09-20. Please provide the PDFs using the suggested filenames; figures and original abstracts can then be added to the existing pages.

Suggested filename Paper Link Material still needed
031.pdf Assessing Domain Generalization of Semantic Segmenters with Synthetic Data Source Full PDF, original abstract, and two figures with captions.
078.pdf mSOP-765k - A Benchmark For Multi-Modal Structured Output Predictions Source Full PDF and at least one additional figure; abstract and one project illustration are available.
085.pdf Fréchet Inception Distance is Failing to Preserve Rank Consistency for Synthetic Out-of-Distribution Samples Source Full PDF, original abstract, and two figures with captions.
089.pdf Parameterized Adverse Lens Corruptions to Probe Model Robustness to Optical Tolerances Source Full PDF, original abstract, and two figures with captions. An indexed abstract excerpt was found, but not a verified complete abstract.
095.pdf We need theory-grounded explainability for abstract tasks to quantify model bias Source Full PDF, original abstract, and two figures with captions.
096.pdf Combining Finetuning and RAG for Structured Output Prediction on a Multimodal High Velocity Retail Dataset Source Full PDF, original abstract, and two figures with captions.
097.pdf Unsupervised Visual Concept Bottlenecks for Interpretable Bioimage Classification Source Full PDF, original abstract, and two figures with captions.

The 2023 domain-generalization entry links to a workshop homepage rather than a paper. Most other downloads returned HTTP 403. A public landing page or search excerpt alone was not counted as access to the full PDF.

Shared preprints

The workshop entries “Robust Models are less Over-Confident” (ICML 2022) and “Data-Efficient Visual Concept Bottleneck Models” (CVPR 2025 workshop) currently use the linked public preprints also used by their later full-conference entries. Separate workshop PDFs are welcome if their abstract or figures differ.

Filling the gaps

Each page is stored in _publications/. Add the verified abstract, abstract_source, and two or three figures entries, then set content_status: complete. Put optimized figure images in images/publications/details/ and record their source pages and crop coordinates in _data/publication-review.yaml.