Human skill, in motion.

The wall combines six open egocentric video excerpts, four licensed stock clips and fifteen existing DVIDIA illustrations. These are examples of human activity, not a live recording feed or a demonstration of model performance. Desktop and mobile each show twenty-five distinct clips in five rows of five. The action names and categories are editorial descriptions, not model predictions, reviewed dataset labels or measured confidence scores.

Upload card — hands in frame

The upload card uses the 0–3 second shirt-folding excerpt from EgoAnnotate v1, GX010079 by Taher Panbiharwala and Zainab Barwaniwala, licensed under CC BY 4.0. DVIDIA muted, resized, color-adjusted, lightly sharpened, added monochrome film grain and slowed the excerpt to two-thirds speed. The card adds a blue tint, a grain overlay, a legibility gradient and looping. This is a framing example, not a live capture, uploaded user video, or AI analysis result. Edits and asset checksums.

EgoAnnotate v1

EgoAnnotate v1 — Taher Panbiharwala and Zainab Barwaniwala (2026). Licensed under Creative Commons Attribution 4.0 International.

• Onion cutting — GX010072, 0–3 seconds.
• Watermelon cutting — GX010076, 0–3 seconds.
• Starting to fold a shirt — GX010079, 0–3 seconds.
• Folding jeans — GX010081, 0–3 seconds.
• Packing food — GX010087, 0–3 seconds.

The publisher normalized frame rates and removed audio. DVIDIA trimmed, resized, cropped, re-encoded and looped these clean video excerpts in the wall. Publisher license notice.

CoMind

CoMind (ECCV 2026) — Alexey Gavryushin, Dingxi Zhang, Zhao Huang, Alexandros Delitzas, Jiaqi Chen, Ben Ellis, Cedric Zöllner, Manthan Patel, Manuel Kaufmann, Marc Pollefeys and Xi Wang. © 2026 CoMind Team. Licensed under Creative Commons Attribution 4.0 International.

Stirring vegetables — leader-view example, 1–5 seconds. DVIDIA trimmed, resized, cropped, re-encoded, muted and looped this excerpt. Later footage is excluded.

Pexels footage

The following clips are licensed under the Pexels License. They are website illustrations, not an open training dataset. DVIDIA trimmed, cropped, resized, muted, re-encoded and looped them into the wall. Only the handwashing clip is verified first-person footage; the others are complementary close-ups.

• Washing hands — K (@kelly), 2–12 seconds, first-person POV.
• Working garden soil — K (@kelly), 2–12 seconds, side close-up.
• Mixing ingredients — Cezar Bulat, 0–3.5 seconds, side close-up.
• Cleaning a tool — Suraphat Nuea-on, 1–11 seconds, overhead close-up.

The remaining tiles

Cup placement, typing, screw turning, opening a jar, wiping, plugging in, watering a plant, tying a shoelace, using scissors, hammering, turning a book page, hanging a shirt, rinsing a plate, peeling an orange and folding a towel are existing DVIDIA repository illustrations. Some include illustrative hand overlays; these are not verified motion measurements. They are separate from the CC BY dataset excerpts credited above.

The credited creators do not endorse DVIDIA. CC BY rights remain available for the six credited excerpts. Those six clips also appear in DVIDIA's starter skillspaces as short reference examples, with AI draft observations and pending human review. Their thumbnail images were extracted at 0.5 seconds into each excerpt. They are not complete training sets. Source URLs, edits and excerpt checksums.