GTR: Gated Token Recurrence for Efficient Dense Prediction
GTR replaces softmax attention with gated token recurrence for efficient high-resolution dense prediction, achieving 58.9 COCO box AP with low latency.
Published Sep 22, 2026▲ 12 on Hugging FaceCode ★ 50arXiv ↗
Only vote on papers you've read. Sign in with GitHub to vote.
GTR delivers remarkable edge latency and COCO box AP through a custom CUDA kernel and DINOv3 distillation, but its dense prediction claims remain unverified without segmentation benchmarks, and its architecture is unproven once stripped of teacher…
Abstract
Self-attention-based vision backbones perform well on dense prediction, but the quadratic computational cost of global softmax attention limits their efficiency as image resolution increases. We introduce Gated Token Recurrence (GTR), a softmax-free recurrent vision backbone that combines gated linear attention, alternating spatial scan directions, and spatially enhanced SwiGLU blocks. GTR is distilled from a detection-specialized DINOv3 teacher using only final-layer patch-token alignment through a linear projection and squared $\ell_2$ loss, without masked-token prediction or intermediate-layer supervision. With Objects365 detector pre-training, GTR-L achieves 58.9 box AP on COCO \texttt{val2017} with 1.908\,ms median batch-one latency under compiled FP16 execution on an RTX~4090. The same backbone also transfers to instance segmentation, pose estimation, oriented detection, semantic segmentation, and monocular depth estimation. In an isolated kernel benchmark, our specialized chunkwise CUDA operator is $4.0\times$ faster than FLA v0.5.0 at 1.6K tokens on RTX~4090. TensorRT deployment on DRIVE AGX Thor achieves 2.282--8.769\,ms median batch-one latency across the evaluated models. These results show that recurrent token mixing can provide an efficient alternative to global softmax attention for high-resolution dense prediction and edge deployment. Project page: https://intellindust-ai-lab.github.io/projects/GTR/