CMU and Meta Unveil TrackEverything, a 3D Point Tracker for 1000+ Frame Long Videos
AdamWHarley · x · 2026-09-26
Researchers from Carnegie Mellon University and Meta introduce TrackEverything, a dense 3D point tracker that tracks all points across all frames of long videos (1000+ frames).
- Key idea: tie computation to unique 3D scene content rather than redundant 2D pixels, by de-duplicating 3D scene representations.
- The video is represented as persistent 3D scene tracks in world coordinates, which naturally handles camera motion; per-window static/dynamic classification keeps renderings clean.
- The project page shows extensive dense-tracking visualizations on full-length sequences; the paper is on arXiv and code is coming soon.
- Authors include Ayush Jain, Adam W. Harley, Katerina Fragkiadaki, and Jakob Engel.
More from Multimodal
- invideo lets you generate custom video effects by describing them in chat, with 90+ built-in presets free — LudovicCreator · 2026-09-26
- Tiny image models that respond faster than a blink unlock new cheap use cases — airesearch12 · 2026-09-26
- Krea2 Turbo 2-step distill LoRA hits 4.2x faster denoising at near-teacher quality — TimeTruth2490 · 2026-09-26
- Entire motion design video is pure code, zero After Effects — dev open-sources the Claude prompt template — EricBuess · 2026-09-26
- viggle-turbo matches original quality on most prompts, now with ComfyUI support — init-5 · 2026-09-26
- Blender agents meet fal's H3 Max: 3D previs turns into photoreal video — gorkem · 2026-09-26