standard criteria (sharpness

ML

• always-on **wearable cameras** that passively capture your day (Microsoft **SenseCam**, **Narrative Clip**, **Google Clips**, Memoto) → a firehose of thousands of images/day

• **SenseCam** (Microsoft Research, **Hodges et al. 2006**) is the seminal device: a small chest-worn camera with a fish-eye lens that fires **automatically from onboard sensors** (accelerometer, ambient-light, and passive-infrared body-heat triggers) — no shutter press — grabbing ~2,000–3,000 images a day. Tellingly, it was built and studied less as a *camera* than as a **memory prosthesis**: clinical trials found that reviewing SenseCam image streams helped patients with amnesia re-consolidate **autobiographical memory** far better than a written diary. That reframes the wearable lifelog as a tool for *recall* rather than photography — and it is exactly what spawns the big-data **curation, summarization, retrieval, and privacy** problems below.

• the big-data problem they *create*: **curation, summarization, retrieval, and privacy** at personal-archive scale (ties to auto-curation / retrieval above)

• selection becomes the product: **what to keep** — Google Clips ran on-device ML to grab candid moments (the blind / anticliché camera idea, automated, below)

• **privacy & ethics**: bystander consent, always-on recording, lifelog security (→ Human factors ethics; adjacent fields)

• bright targets (the Moon, planets) are blurred and wobbled by atmospheric **seeing**; instead of one exposure, shoot **thousands of fast frames** (usually video) and **keep only the sharpest few %** — the brief moments when the turbulence settles

• a **big-data selection** problem: score every frame by a **sharpness metric**, **rank**, keep the best, then **align + stack** the survivors (1/√N again) → ground-based resolution approaching the telescope's diffraction limit, cheaply (a poor-man's adaptive optics)

• ties to **auto curation** above (rank-and-select from a huge frame set) and to **Denoising by averaging** (stack the keepers, Multiple exposure); refs: Law et al. 2006

• **Photobios** (the *Exploring Photobios* project of **Ira Kemelmacher-Shlizerman** and **Steve Seitz**): face photo collections **over time** — align many portraits of one person to a common frame and order them → a smooth **time-lapse of a face aging** (Kemelmacher-Shlizerman et al. 2011). A big-collection instance of [[Morphing]] + alignment, where the **data** (not a model) supplies the in-betweens; kin to [[#Photo tourism]] (a collection becomes an experience).