ANIMA FOXHOUND — Wave 10 WARDOG

Paper: Learning Motion Blur Robust Vision Transformers for Real-Time UAV Tracking Codename: FOXHOUND ArXiv: https://arxiv.org/abs/2407.05383 Reference repo: https://github.com/wuyou3474/BDTrack Focus: UAV/drone defense for Shenzhen Robot Fair

Status

FOXHOUND is in conditional-go bootstrap status.

• The paper is real and readable from the bundled PDF.
• The public GitHub repo exists, but the implementation is not actually released yet.
• Public benchmark datasets are identifiable, but no shared-volume copies were found during the initial audit.
• The paper reports plausible benchmark numbers, but the full training recipe is partially delegated to Aba-ViTrack.

This means the project is buildable, but reproducibility carries explicit risk and should remain visible in all planning documents.

What FOXHOUND Builds

FOXHOUND adapts BDTrack into an ANIMA-ready module with:

• a ViT tracker scaffold grounded in the paper's MBRV and DEEM ideas,
• YOLO26-compatible proposal hooks for UAV-defense deployment,
• dual-backend runtime selection (mlx, cuda, cpu),
• a full ANIMA PRD suite in prds/ and granular execution tasks in tasks/.

Quick Start

uv pip install -e ".[dev]"
python -m anima_foxhound --describe-paper
python -m anima_foxhound --backend auto
python -m anima_foxhound --backend mlx
python -m anima_foxhound --backend cuda

Build Pack

• PRD.md: top-level execution brief and verification verdict
• ASSETS.md: weights, datasets, and reproducibility inventory
• prds/: 7 ANIMA build PRDs
• tasks/: granular implementation tasks and build order
• NEXT_STEPS.md: execution ledger and red flags

Code Layout

src/anima_foxhound/
  __main__.py          CLI entrypoint
  config.py            TOML-backed runtime configuration
  device.py            backend detection and capability reporting
  paper.py             paper facts, metrics, and reproducibility flags
  blur.py              motion blur simulation helpers
  exit_policy.py       DEEM-style cumulative exit logic
  yolo26_adapter.py    YOLO26 integration hooks
  bootstrap_model.py   paper-grounded bootstrap orchestration
  runtime.py           module summary and environment audit

License

Research use only. Public benchmark and pretrained asset licenses remain governed by their original owners.