T6D-Direct: Transformers for Multi-object 6D Pose Direct Regression

6D pose estimation is the task of predicting translation and orientation objects in a given input image, which crucial prerequisite for many robotics augmented reality applications. Lately, Transformer Network architecture, equipped with multi-head self-attention mechanism, emerging to achieve state-of-the-art results computer vision tasks. DETR, Transformer-based model, formulated object detection as set prediction problem achieved impressive without standard components like region interest pooling, non-maximal suppression, bounding box proposals. In this work, we propose T6D-Direct, real-time single-stage direct method transformer-based architecture built on DETR perform multi-object estimation. We evaluate performance our YCB-Video dataset. Our achieves fastest inference time, accuracy comparable methods.