v1.20.1

v1.20.1 (2025-03-12)

Bug Fixes

• Correct arg for weighted score in cluster (658ef95)

---

Detailed Changes: v1.20.0...v1.20.1

v1.20.0

v1.20.0 (2025-03-12)

Bug Fixes

• Remove cuda in cluster vits (d449a86)

Chores

• Clean up imports (18da1aa)

Continuous Integration

• Only run pytest on push to main branch (fcfb28d)

Documentation

• Updated workflow diagram (91ebe6a)

Features

• Add weight_vits option to weight the scores from the detection model in the vits classification model (9bffe3d)

• Added feature merge (90d182d)

• Added hdbscan algorithm choice for clustering (40b9ba2)

• Added hdbscan algorithm choice for clustering (722da61)

• Added min-sample-size argument to allow for parameter sweeps (2cf9771)

• Rename to weighted_score and add back in the noise reassignment for higher coverage (0a361a0)

Performance Improvements

• Improved cluster coverage, weighted classification scores, and more options for running cluster sweeps (a94e4a9)

Performance

• Better handling of noise cluster and merging similar clusters. This should improve cluster coverage and generate somewhat larger clusters with foundation models.

Features

• new arg to sdcat cluster --algorithm default "best"; prims_kdtree or boruvka_kdtree may be worth trying
• new arg to sdcat cluster --min-sample-size which was only supported in the .ini file
• new arg to sdcat cluster --weighted-score which will weight the classification score with the detection score from the ViTS models through multiplication

---

Detailed Changes: v1.19.1...v1.20.0

v1.19.1

v1.19.1 (2025-02-26)

Bug Fixes

• Correct handling of bounded end image (f510e16)

---

Detailed Changes: v1.19.0...v1.19.1

v1.19.0

v1.19.0 (2025-02-26)

Features

• Added softcluster (#19, cfaf784)

• perf: better defaults for finer-grained clustering with google model

• feat: added soft clustering for leaf method only

• fix: remove default as this overrides what is in the .ini file

• perf: add batch size as command option --batch-size; default is 32 but best size depends on GPU/model memory

• fix: correct args for multiproc

• perf: combine soft/fuzzy and cosine sim

• docs: update workflow diagram with soft/fuzzy algorithm

• fix: handle models that only output top 1

• fix: only capture top 2 classes and scores

• chore: merged changes from main

---

Detailed Changes: v1.18.2...v1.19.0

v1.18.2

v1.18.2 (2025-02-20)

Bug Fixes

• Only capture top 2 classes and scores (9b85463)

---

Detailed Changes: v1.18.1...v1.18.2

v1.18.1

v1.18.1 (2025-02-20)

Bug Fixes

• Handle models that only output top 1 and default to cuda if available if not specified for clustering (164480a)

---

Detailed Changes: v1.18.0...v1.18.1

v1.18.0

v1.18.0 (2025-02-20)

Features

• Add support for --save-roi --roi-size (#18, 9a801ac)

Added --save-roi and --roi-size options to sdcat detect. This saves the crops in a location compatible with the clustering stage, but can also be used outside of sdcat. Data saved to crops

├── det_filtered # The filtered detections from the model
├── crops # Crops of the detections

• Trigger release for --save-roi (8240a74)

---

Detailed Changes: v1.17.0...v1.18.0

v1.17.0

v1.17.0 (2025-02-07)

Build System

• Relaxed requirements for compatibility with mbari-aidata since these are often used together (8bf55e3)

Features

• Trigger release to pypi with latest deps (2490823)

---

Detailed Changes: v1.16.3...v1.17.0

v1.16.3

v1.16.3 (2025-01-27)

Build System

• Updated poetry lock (b9a04e6)

Performance Improvements

• Bump sahi to support YOLOv11 (d36b494)

---

Detailed Changes: v1.16.2...v1.16.3

v1.16.2

v1.16.2 (2025-01-14)

Performance Improvements

• Better handling of cuda devices by id across both detection and clustering commands with --device cuda:0 (ae8e395)

---

Detailed Changes: v1.16.1...v1.16.2