DiffVL: Diffusion‑Based Visual Localization Using GPS Denoising
DiffVL refines noisy GPS with visual BEV cues, reaching sub‑meter accuracy without any HD maps by using freely available SD map data and standard GPS hardware. Read more: getnews.me/diffvl-diffusion-based-v... #diffvl #visuallocalization #gps
Tracking results of iLoc. This plot shows the localization results in our urban dataset. The yellow line represents the trajectory of the reference map, and the red line represents the tracking trajectory. iLoc will first localize itself in the reference map by proposed fast-seq matching to find the best match of the reference frame. In the tracking procedure, iLoc continuously matches the observed panorama with the local sliding window to find the best match.
A transformer-based, coarse-to-fine #VisualLocalization pipeline that extracts stable, illumination-invariant descriptors for place recognition, achieves sub-meter global relocalization in under 0.5 s across 2 km and provides real-time tracking.
ieeexplore.ieee.org/document/108...
📢 Call for 8-page papers & 4-page abstracts: Opens on the 16th of May!
🏆 Challenge coming soon—stay tuned!
#CroCoDL #AI #ICCV #AR #VisualLocalization
🤖 @marcpollefeys.bsky.social @mihaidusmanu.bsky.social @hermannblum.bsky.social @iccv.bsky.social @ethz.ch
🧵 2/2
🧵6/9
Gaussian Splatting Feature Fields for (Privacy-Preserving) Visual Localization
We combine 3D Gaussian Splatting & feature fields for accurate, privacy-friendly localization. Pietrantoni et al. #VisualLocalization
Abstract: tinyurl.com/ye27w5zt
