Abstract:
End-to-end Object Detection with Transformer (DETR) performs object detection with Transformer and achieves comparable performance with two-stage object detection like Faster-RCNN. However, DETR needs huge computational resources for training and inference due to the high-resolution spatial inputs. In this paper, a novel variant of transformer named Adaptive Clustering Transformer (ACT) has been proposed to reduce the computation cost for high-resolution input. ACT clusters the query features adaptively using Locality Sensitive Hashing (LSH) and approximates the query-key interaction using the prototype-key interaction. ACT can reduce the quadratic O(N^2) complexity inside self-attention into O(NK) where K is the number of prototypes in each layer. ACT can be a drop-in module replacing the original self-attention module without any training. ACT achieves a good balance between accuracy and computation cost (FLOPs). The code is available as supplementary for the ease of experiment replication and verification.
