Dense Prediction Transformer

Dense prediction transformers are a novel approach leveraging the power of transformer architectures for tasks requiring pixel-wise predictions in computer vision. Current research focuses on applying these models to challenges like 3D semantic occupancy prediction, depth estimation from monocular vision (including fusion with radar data), and scale estimation in visual odometry. These advancements improve the accuracy and efficiency of various computer vision applications, particularly in autonomous driving and robotics, by offering superior performance compared to traditional convolutional neural networks. The development of memory-efficient training strategies and effective multi-modal fusion techniques are key areas of ongoing investigation.

Papers

November 19, 2023

SOccDPT: Semi-Supervised 3D Semantic Occupancy from Dense Prediction Transformers trained under memory constraints
Aditya Nalgunda Ganesh
Semantic Segmentation Traffic Datasets Memory Constraint Semantic Occupancy 3D Semantic Occupancy Dense Prediction Transformer

November 4, 2022

RCDPT: Radar-Camera fusion Dense Prediction Transformer
Chen-Chou Lo, Patrick Vandewalle
Vision Transformer Depth Estimation Convolutional Backbone Dense Prediction Transformer

October 4, 2022

Dense Prediction Transformer for Scale Estimation in Monocular Visual Odometry
André O. Françani, Marcos R. O. A. Maximo
Monocular Vision Monocular Visual Odometry Scale Estimation Dense Prediction Transformer

Dense Prediction Transformer

Papers

SOccDPT: Semi-Supervised 3D Semantic Occupancy from Dense Prediction Transformers trained under memory constraints

RCDPT: Radar-Camera fusion Dense Prediction Transformer

Dense Prediction Transformer for Scale Estimation in Monocular Visual Odometry