Exploring Plain Vision Transformer Backbones for Object Detection

TL;DR

This paper investigates the use of plain, non-hierarchical Vision Transformer backbones for object detection, demonstrating competitive performance with minimal architectural modifications and simple feature pyramid design.

Contribution

It introduces ViTDet, a plain-backbone detector using ViT pre-trained as MAE, achieving competitive results without hierarchical backbones or complex attention mechanisms.

Findings

Plain ViT backbones can be effectively used for object detection.

A simple feature pyramid from a single-scale feature map suffices.

Using window attention without shifting is adequate for strong performance.

Abstract

We explore the plain, non-hierarchical Vision Transformer (ViT) as a backbone network for object detection. This design enables the original ViT architecture to be fine-tuned for object detection without needing to redesign a hierarchical backbone for pre-training. With minimal adaptations for fine-tuning, our plain-backbone detector can achieve competitive results. Surprisingly, we observe: (i) it is sufficient to build a simple feature pyramid from a single-scale feature map (without the common FPN design) and (ii) it is sufficient to use window attention (without shifting) aided with very few cross-window propagation blocks. With plain ViT backbones pre-trained as Masked Autoencoders (MAE), our detector, named ViTDet, can compete with the previous leading methods that were all based on hierarchical backbones, reaching up to 61.3 AP_box on the COCO dataset using only ImageNet-1K pre-training. We hope our study will draw attention to research on plain-backbone detectors. Code for ViTDet is available in Detectron2.