Multi-dataset Pretraining: A Unified Model for Semantic Segmentation

TL;DR

This paper introduces a unified multi-dataset pretraining framework for semantic segmentation that leverages fragmented annotations across datasets, improving transferability and performance with less pretraining data.

Contribution

It proposes a novel pixel-to-prototype contrastive loss and pixel-to-class sparse coding to effectively utilize multiple datasets for pretraining.

Findings

Outperforms ImageNet pretraining by a significant margin.

Requires less than 10% of samples for pretraining.

Achieves superior performance on several benchmarks.

Abstract

Collecting annotated data for semantic segmentation is time-consuming and hard to scale up. In this paper, we for the first time propose a unified framework, termed as Multi-Dataset Pretraining, to take full advantage of the fragmented annotations of different datasets. The highlight is that the annotations from different domains can be efficiently reused and consistently boost performance for each specific domain. This is achieved by first pretraining the network via the proposed pixel-to-prototype contrastive loss over multiple datasets regardless of their taxonomy labels, and followed by fine-tuning the pretrained model over specific dataset as usual. In order to better model the relationship among images and classes from different datasets, we extend the pixel level embeddings via cross dataset mixing and propose a pixel-to-class sparse coding strategy that explicitly models the pixel-class similarity over the manifold embedding space. In this way, we are able to increase intra-class compactness and inter-class separability, as well as considering inter-class similarity across different datasets for better transferability. Experiments conducted on several benchmarks demonstrate its superior performance. Notably, MDP consistently outperforms the pretrained models over ImageNet by a considerable margin, while only using less than 10% samples for pretraining.