PhenoTypic — third-party component notices
==========================================

PhenoTypic is licensed under Apache-2.0 (see LICENSE). It optionally depends on
the third-party GPU-detection components below. PhenoTypic does NOT redistribute
any model weights: model weights are downloaded by the user from the upstream
source under that model's own license, which the user must accept.

- SAM2 (segment-anything-2), Meta — code & weights under Apache-2.0.
  Upstream:  https://github.com/facebookresearch/sam2
  License:   licenses/sam2-Apache-2.0.txt

- micro-sam (computational-cell-analytics) — code under the MIT License;
  checkpoints are distributed via its own channels under their respective terms.
  Upstream:  https://github.com/computational-cell-analytics/micro-sam
  License:   licenses/micro-sam-LICENSE.txt

- SAM3 (segment-anything-3), Meta — code via `transformers` (Apache-2.0);
  weights under the SAM License (commercial-OK, GATED on Hugging Face).
  Upstream:  https://huggingface.co/facebook/sam3
  License:   licenses/sam3-SAM-License.txt

- DINOv2, Meta — backbone for DinoSam2Detector; Apache-2.0, ungated.
  Upstream:  https://github.com/facebookresearch/dinov2
  License:   licenses/dinov2-Apache-2.0.txt

- DINOv3, Meta — frozen backbone for Insid3Detector (native) and the optional
  DINOv3 path of DinoSam2Detector / FssDinoDetector; weights under the DINOv3
  License (commercial-OK, redistributable only under that license, GATED on
  Hugging Face). Per the DINOv3 License §1.b.i, PhenoTypic prominently displays
  "Built with DINOv3" wherever DINOv3 features are produced.
  Upstream:  https://huggingface.co/facebook/dinov3-vitb16-pretrain-lvd1689m
  License:   licenses/dinov3-License.txt

- INSID3 (visinf/INSID3) — training-free in-context (one-shot) semantic
  segmentation on a frozen DINOv3 backbone; code under Apache-2.0. PhenoTypic's
  Insid3Detector is a clean-room reimplementation of its positional-bias-removal
  + prototype-matching method, attributed here; no upstream code is vendored.
  Upstream:  https://github.com/visinf/INSID3
  License:   licenses/insid3-Apache-2.0.txt

- FSSDINO (Zakir & Ho, "Revealing the Semantic Selection Gap in DINOv3 through
  Training-Free Few-Shot Segmentation", arXiv:2602.07550) — few-shot semantic
  segmentation (class-specific prototypes + Gram-matrix refinement) on a frozen
  DINO backbone. The reference code repository carries NO license (all rights
  reserved) and is NOT vendored; PhenoTypic's FssDinoDetector is a CLEAN-ROOM
  reimplementation FROM THE PAPER ONLY. The paper is licensed CC BY-NC-SA and is
  attributed here. No FSSDINO code is bundled or redistributed.
  Paper:     https://arxiv.org/abs/2602.07550

Gated foundation-model weights (SAM3, DINOv3) are never bundled; their downloads
are gated behind an explicit license-acceptance step (see
PHENOTYPIC_ACCEPT_MODEL_LICENSE and the require_license_acceptance hook in
src/phenotypic/detect/nn/_checkpoint_manager.py).
