**Note** from Bead: Image Pose Uxui · [canonical source](https://redfish.acequia.io/guerin/.agents/74c30681-2b0c-48b2-b2a8-9e21190a4926/2026-07-03/notes/00-fft-pose-estimation-directive.md) · session 2026-07-03 · discussion: Talk: Image Pose Uxui
## Directive Stephen requests exploration of **FFT (Fast Fourier Transform)** techniques for image pose estimation as an alternative or complement to the current work-queue approach (manual snapshot-view + Campanile pinning).
## Context - Mounted bead: `74c30681` (image-pose-uxui, 3D pose calibrator) - Previous work: manual interaction workflow (steps 0–9 in WORK-QUEUE.md) - Current step: exploratory research into FFT-based pose methods - Reference chat: to be imported from Claude share URL
## Questions to explore 1. **FFT phase correlation** — use FFT to compute phase shift between image and reference template; applicable to rotation/translation estimation? 2. **Frequency-domain features** — extract pose-invariant or pose-dependent features in frequency space? 3. **Template matching via FFT** — fast cross-correlation for coarse pose hypotheses? 4. **Hybrid approach** — use FFT for rapid pose initialization, then refine with Levenberg-Marquardt (per `#nuke-geo-camera` solver)?
## Integration points - Existing solver: `solver.js` (currently empty; hooks to `#nuke-geo-camera` camera-calibrator) - Existing mounts: `#nuke-geo-camera` (bundle adjustment, Levenberg-Marquardt), taos-engine 3D render - Existing UI: calibrate-3d.html (will add FFT-based pose hypothesis layer)
## Next steps 1. Capture shared chat into bead `chats/` 2. Extract FFT approach(es) into exploratory `notes/` 3. Prototype FFT phase-correlation in `repo/fft-pose-prototype.html` 4. Measure speed/accuracy vs manual workflow 5. Decide: replace snapshot-pose, supplement it, or defer to follow-up