Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions

TL;DR

This paper introduces Pyramid Vision Transformer (PVT), a convolution-free backbone that enhances dense prediction tasks by combining Transformer advantages with high-resolution outputs and reduced computation, outperforming traditional CNNs.

Contribution

The paper proposes PVT, a novel Transformer-based backbone that supports high-resolution dense predictions and reduces computation, bridging the gap between CNNs and Transformers for vision tasks.

Findings

PVT achieves higher AP on COCO compared to ResNet50-based models.

PVT effectively boosts performance in object detection and segmentation tasks.

PVT demonstrates competitive results with fewer parameters.

Abstract

Although using convolutional neural networks (CNNs) as backbones achieves great successes in computer vision, this work investigates a simple backbone network useful for many dense prediction tasks without convolutions. Unlike the recently-proposed Transformer model (e.g., ViT) that is specially designed for image classification, we propose Pyramid Vision Transformer~(PVT), which overcomes the difficulties of porting Transformer to various dense prediction tasks. PVT has several merits compared to prior arts. (1) Different from ViT that typically has low-resolution outputs and high computational and memory cost, PVT can be not only trained on dense partitions of the image to achieve high output resolution, which is important for dense predictions but also using a progressive shrinking pyramid to reduce computations of large feature maps. (2) PVT inherits the advantages from both CNN and Transformer, making it a unified backbone in various vision tasks without convolutions by simply replacing CNN backbones. (3) We validate PVT by conducting extensive experiments, showing that it boosts the performance of many downstream tasks, e.g., object detection, semantic, and instance segmentation. For example, with a comparable number of parameters, RetinaNet+PVT achieves 40.4 AP on the COCO dataset, surpassing RetinNet+ResNet50 (36.3 AP) by 4.1 absolute AP. We hope PVT could serve as an alternative and useful backbone for pixel-level predictions and facilitate future researches. Code is available at https://github.com/whai362/PVT.