AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Translating Images to Road Network: A Non-Autoregressive Sequence-to-Sequence Approach

Jiachen Lu, Renyuan Peng, Xinyue Cai, Hang Xu, Hongyang Li, Feng Wen, Wei Zhang, Li Zhang

ICCV 2023 · Architectures and efficiency › Attention and transformers · approach: State space and recurrent sequence models · 13 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
decouple autoregressive dependency into semi-autoregressive generationRoadNetTransformer · nuScenes multi-camerareachability F1 / FPS: R-F 61.4 -> 67.1; FPS 0.1 -> 0.6 (6.0x)Improved
use non-autoregressive masked iterative generation after SARRoadNetTransformer · nuScenes multi-cameraFPS: 0.1 -> 4.7 (47x); R-F 61.4 -> 66.0Improved
NAR instead of SAR (ablation)RoadNetTransformer · nuScenes multi-camerareachability F1: 67.1 -> 66.0Hurt
add Intra-seq self-attention in the parallel-seq decoder (ablation)NAR-RNTR · nuScenes multi-camerareachability F1: 56.3 -> 66.0Improved
increase key-point decoder layers from 3 to 6 (ablation)NAR-RNTR · nuScenes multi-cameralandmark F1: 56.1 -> 60.2Improved
increase parallel-seq decoder layers from 3 to 6 (ablation)NAR-RNTR · nuScenes multi-cameralandmark F1 / FPS: L-F 60.2 -> 61.3; FPS 4.7 -> 3.0Mixed
raise training mask ratio from 50% to 90% (ablation)NAR-RNTR · nuScenes multi-camerareachability F1: 51.3 -> 66.0Improved
increase inference iterations from 3 to 6 (ablation)NAR-RNTR · nuScenes multi-camerareachability F1: 66.0 -> 66.1; FPS 4.7 -> 2.8No clear effect
use RoadNet Sequence transformers instead of STSU/TPLRfront-camera road network extraction · nuScenes PON valmean F1: 58.2 -> 59.3 (AR); 61.6 (SAR)Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy