{"paper":{"title":"Weakly Supervised Segmentation as Semantic-Based Regularization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Differentiable fuzzy logic encodes weak annotations as continuous constraints to fine-tune SAM and generate higher-quality pseudo-labels for segmentation.","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Andrei-Bogdan Florea, Jaron Maene, Stefano Colamonaco","submitted_at":"2026-05-13T15:33:27Z","abstract_excerpt":"Weakly supervised semantic segmentation (WSSS) trains dense pixel-level segmentation models from partial or coarse annotations such as bounding boxes, scribbles, or image-level tags. While recent work leverages foundation models such as the Segment Anything Model (SAM) to generate pseudo-labels, these approaches typically depend on heuristic prompt choices and offer limited ways to incorporate prior knowledge or heterogeneous labels. We address this gap by taking a neurosymbolic perspective: integrating differentiable fuzzy logic with deep segmentation models. Weak annotations and domain-speci"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experiments on Pascal VOC 2012 and the REFUGE2 optic disc/cup segmentation dataset show that our logic-guided fine-tuning yields higher-quality pseudo-labels, leading to state-of-the-art segmentation accuracy that often exceeds densely supervised baselines.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That differentiable fuzzy logic can reliably encode heterogeneous weak annotations and domain priors as continuous constraints without introducing systematic biases or requiring dataset-specific tuning that undermines generalization.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Differentiable fuzzy logic constraints fine-tune SAM to generate higher-quality pseudo-labels, enabling a second-stage model to reach state-of-the-art weakly supervised segmentation on Pascal VOC and REFUGE2, sometimes beating dense supervision.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Differentiable fuzzy logic encodes weak annotations as continuous constraints to fine-tune SAM and generate higher-quality pseudo-labels for segmentation.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"c83c7d6139da78bb9a23c4415c78d85fc6fa18b6c111444f5e353e0cd4d457b3"},"source":{"id":"2605.13674","kind":"arxiv","version":1},"verdict":{"id":"1b68ca72-283e-47f7-9b4b-84b0645277b7","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T20:24:40.043271Z","strongest_claim":"Experiments on Pascal VOC 2012 and the REFUGE2 optic disc/cup segmentation dataset show that our logic-guided fine-tuning yields higher-quality pseudo-labels, leading to state-of-the-art segmentation accuracy that often exceeds densely supervised baselines.","one_line_summary":"Differentiable fuzzy logic constraints fine-tune SAM to generate higher-quality pseudo-labels, enabling a second-stage model to reach state-of-the-art weakly supervised segmentation on Pascal VOC and REFUGE2, sometimes beating dense supervision.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That differentiable fuzzy logic can reliably encode heterogeneous weak annotations and domain priors as continuous constraints without introducing systematic biases or requiring dataset-specific tuning that undermines generalization.","pith_extraction_headline":"Differentiable fuzzy logic encodes weak annotations as continuous constraints to fine-tune SAM and generate higher-quality pseudo-labels for segmentation."},"references":{"count":60,"sample":[{"doi":"","year":2018,"title":"In: Proceedings of the IEEE conference on computer vision and pattern recognition","work_id":"07c39fba-13a9-48dc-89be-389f8269ab14","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Segdiff: Image segmentation with diffusion proba- bilistic models","work_id":"3afa57d7-f737-46ae-b059-5598f4c3b281","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Artificial Intelligence303, 103649 (2022)","work_id":"f147334c-9067-41ef-8ed5-110985f58875","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"In: 2025 International Joint Conference on Neural Networks (IJCNN)","work_id":"41d6baa3-d72e-4627-9273-b596d4cf3f1d","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"TransUNet: Transformers Make Strong Encoders for Medical Image Segmentation","work_id":"ffa2ac60-2755-4390-9f66-07815aa6cb27","ref_index":5,"cited_arxiv_id":"2102.04306","is_internal_anchor":true}],"resolved_work":60,"snapshot_sha256":"869af3f02837e6c72ba8d66b021b09e4e7fee61b15eb3c15ff738d00ab069131","internal_anchors":2},"formal_canon":{"evidence_count":2,"snapshot_sha256":"cd6694a39bc0772f036394e9ed5ad9d3593ad3cfef7f1cf535b665a00c13da6c"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}