cite: add arXiv eprint to the V2N BibTeX

add arXiv paper links (nav, hero, resources) now that the V2N paper is public

viewer: load clip and prediction audio as blobs to bypass range-locked LFS redirect that blocks iOS playback

viewer: tap-to-play overlay and webkit-playsinline so the video starts on mobile when autoplay is blocked

viewer: playback speed 0.25/0.5/1x on video and prediction audio; unify citation to last-first

Card gradient to blue (matches page theme)

Emoji: 🎹 -> 👀 (V2N watches the keys)

Open external links in a new tab (fix GitHub iframe block on HF Space)

Fix PPAN reference author name: Zivanovic (no diacritic, matching arXiv/IJCAI and the paper)

Author credit: mark equal contribution (Kim, Sohn) and map affiliations

Even, fixed-width result columns with a tighter width and larger type

Unify the results table with the ablation tables: sans tabular-nums, tighter padding, no mono

Clearer demo heading: watch it transcribe note by note

Animation: velocity read at the onset (not across key hold), clear x3 blocks label; cite S2S at first mention; clearer physical-key-release card

Make the offset story precise and intuitive: audio misses key release not sound end, and gloss velocity as how hard each key is struck

Say a note keeps sounding rather than rings on, less bell-like for piano

Show a filmstrip poster frame before the pipeline animation scrolls into view

Add an animated 3Blue1Brown-style pipeline in How it works: frames to ResNet features to Conformer mixing to four color-matched heads to decoded MIDI, looping in view

Two-line hero lede and single CTA, color-coded compact head cards matching the figure, numbered baseline references with jump links, window sizes on the multi-frame ablation, and lighter demo copy

Cite the baselines with links to their papers, plainer section headings, rename the resources eyebrow, keep the hero lede noun phrase intact

Tighten the copy across hero, demo, why-video, and results, and wrap the table caption so no line is left an orphan

Kill line orphans with text-wrap balance/pretty; bold V2N (best in every column) in the main table and the tied-best ablation cell, with a bold-means-best note

Add multi-task and multi-frame ablation section, cumulative-metric note, author and institution links, disabled paper badge; tighten the complete footnote and italicize the four attributes

Footnote defining what complete means (all MIDI note attributes, not fingering or expression); switch display face to Geist

Fix the real velocity-color bug: mix() returned an rgb() string that hex() reparsed to near-black when nested, so soft notes rendered black; return hex instead

Fix velocity color: opaque played notes so overlaps stop muddying, single clean blue ramp, GT outline on top, legend generated from the same colors

Make predicted velocity legible: percentile-normalized color, opacity, and bar-height ramp in the piano roll

Plant the between/V2N wordplay subtly, bold the two words in the demo lead

Ground page in the paper: honest demo, key-mechanics and architecture figures, real baseline table across PianoVAM and R3

Multi-piece selector, fixed audio sync, bigger legend and fonts, speaker icons, Between tagline

Apple-style redesign: GT vs prediction, velocity color, audio A/B

Interactive landing page with live PianoVAM demo

V2N landing page

initial commit