GenMask: Adapting DiT for Segmentation via Direct Mask Generation

TL;DR

GenMask introduces a unified generative approach for segmentation by directly training a DiT model to generate masks and images, overcoming the limitations of indirect feature retrieval methods.

Contribution

It proposes a novel training strategy with timestep sampling for binary masks, enabling DiT to generate segmentation masks directly without specialized feature extraction pipelines.

Findings

Achieves state-of-the-art results on segmentation benchmarks.

Demonstrates effective joint training of mask and image generation.

Quantifies the impact of each component through ablation studies.

Abstract

Recent approaches for segmentation have leveraged pretrained generative models as feature extractors, treating segmentation as a downstream adaptation task via indirect feature retrieval. This implicit use suffers from a fundamental misalignment in representation. It also depends heavily on indirect feature extraction pipelines, which complicate the workflow and limit adaptation. In this paper, we argue that instead of indirect adaptation, segmentation tasks should be trained directly in a generative manner. We identify a key obstacle to this unified formulation: VAE latents of binary masks are sharply distributed, noise robust, and linearly separable, distinct from natural image latents. To bridge this gap, we introduce timesteps sampling strategy for binary masks that emphasizes extreme noise levels for segmentation and moderate noise for image generation, enabling harmonious joint training. We present GenMask, a DiT trains to generate black-and-white segmentation masks as well as colorful images in RGB space under the original generative objective. GenMask preserves the original DiT architecture while removing the need of feature extraction pipelines tailored for segmentation tasks. Empirically, GenMask attains state-of-the-art performance on referring and reasoning segmentation benchmarks and ablations quantify the contribution of each component.