GTR-Depth: metric depth from a single camera at 2.7 ms on DRIVE AGX Thor
retained text
Follow-up to last week's GTR post. This one is all about depth. GTR-Depth predicts absolute depth in meters from a single image, using the same backbone as before: 12 gated linear attention blocks, no softmax attention anywhere.
document fetched 2026-10-04 00:22 · canonical https://reddit.com/r/computervision/comments/1wwptmx/gtrdepth_metric_depth_from_a_single_camera_at_27