FMS$^2$: Unified Flow Matching for Segmentation and Synthesis of Thin Structures

Babak Asadi; Peiyang Wu; Mani Golparvar-Fard; Viraj Shah; Ramez Hajj

cs.CVMar 14, 2026

FMS $^2$ : Unified Flow Matching for Segmentation and Synthesis of Thin Structures

Authors: Babak Asadi, Peiyang Wu, Mani Golparvar-Fard, Viraj Shah, Ramez Hajj

Organizations: University of Illinois at Urbana-Champaign

Abstract

Segmenting thin structures like infrastructure cracks and anatomical vessels is a task hampered by topology-sensitive geometry, high annotation costs, and poor generalization across domains. Existing methods address these challenges in isolation. We propose FMS $^2$ , a flow-matching framework with two modules. (1) SegFlow is a 2.96M-parameter segmentation model built on a standard encoder-decoder backbone that recasts prediction as continuous image $\rightarrow$ mask transport. It learns a time-indexed velocity field with a flow-matching regression loss and outputs the mask via ODE integration, rather than supervising only end-state logits. This trajectory-level supervision improves thin-structure continuity and sharpness, compared with tuned topology-aware loss baselines, without auxiliary topology heads, post-processing, or multi-term loss engineering. (2) SynFlow is a mask-conditioned mask $\rightarrow$ image generator that produces pixel-aligned synthetic image-mask pairs. It injects mask geometry at multiple scales and emphasizes boundary bands via edge-aware gating, while a controllable mask generator expands sparsity, width, and branching regimes. On five crack and vessel benchmarks, SegFlow alone outperforms strong CNN, Transformer, Mamba, and generative baselines, improving the volumetric metric (mean IoU) from 0.511 to 0.599 (+17.2%) and reducing the topological metric (Betti matching error) from 82.145 to 51.524 (-37.3%). When training with limited labels, augmenting SegFlow with SynFlow-generated pairs recovers near-full performance using 25% of real annotations and improves cross-domain IoU by 0.11 on average. Unlike classical data augmentation that promotes invariance via label-preserving transforms, SynFlow provides pixel-aligned paired supervision with controllable structural shifts (e.g., sparsity, width, branching), which is particularly effective under domain shift.

Figures & tables

Figure 1: FMS 2 Framework. SegFlow recasts segmentation as image $\rightarrow$ mask transport problem, effectively preserving sub-pixel topologies and yielding sharper segmentations than existing approaches. SynFlow is mask $\rightarrow$ image renderer that generates topology-consistent image-mask pairs to alleviate the scarcity of labeled data for SegFlow training.

Figure 2: Architecture and transport view of FMS 2 . (a) Coupled SegFlow (image $\rightarrow$ mask) and SynFlow (mask $\rightarrow$ image) share a U-Net–like backbone and are trained with flow matching: a time-indexed velocity field $v_{\theta}(x_{t},t,\cdot)$ transports noise to images and masks, while multi-scale mask injection steers trajectories to preserve one-pixel topology. (b) Toplogy-Aware Mask Injection (TAMI) used in both flows: in SynFlow it combines lite-SPADE GroupNorm (LSGN) with the boundary-gating module (BGM) to inject mask geometry and focus capacity near thin structures, while in SegFlow it reduces to GN-based modulation. (c) LSGN uses a mask pyramid to generate affine parameters that modulate group-normalized features, whereas BGM builds an edge band via morphological dilate/erode and gates features near crack/vessel interfaces, emphasizing topology-critical regions.

Figure 3: Impact of integration steps in SegFlow on segmentation performance. We report the change (relative to step=1) in mIoU and

\mu^{\mathrm{err}}

across datasets as the number of integration steps varies.

Figure 4: Visual comparison of SegFlow against SOTA methods. Yellow arrows denote missed structures (FNs), green arrows indicate spurious predictions (FPs).

Figure 5: Comparison of real vs. SynFlow-generated mask–image pairs. SynFlow preserves pixel alignment between masks and rendered images while producing realistic textures; additional examples are provided in the supplementary material.

Figure 6: Effect of SynFlow synthesis on label efficiency. SegFlow performance when training with a small portion of real labels (

0.25R

) and progressively adding SynFlow images (

0.25R{+}4S

0.25R{+}16S

0.25R{+}64S

). Here

R

is the full real training set (

R{=}400

for Crack500 and CrackTree260,

R{=}100

for XCAD) and

S{=0.25}R

; the dashed line marks full real supervision at

R

. Left: IoU, right: clDice.

Thin-Structure Segmentation

See all 12 papers

Apr 3, 2026

Few-Shot Distribution-Aligned Flow Matching for Data Synthesis in Medical Image Segmentation

Data heterogeneity hinders clinical deployment of medical image analysis models, and generative data augmentation helps mitigate this issue. However, recent diffusion-based methods that synthesize image-mask pairs often ignore distribution shifts between generated and real images across scenarios, and such mismatches can markedly degrade downstream performance. To address this issue, we propose AlignFlow, a flow matching model that aligns with the target reference image distribution via differentiable reward fine-tuning, and remains effective even when only a small number of reference images are provided. Specifically, we divide the training of the flow matching model into two stages: in the first stage, the model fits the training data to generate plausible images; Then, we introduce a distribution alignment mechanism and employ differentiable reward to steer the generated images toward the distribution of the given samples from the target domain. In addition, to enhance the diversity of generated masks, we also design a flow matching based mask generation to complement the diversity in regions of interest. Extensive experiments demonstrate the effectiveness of our approach, i.e., performance improvement by 3.5-4.0% in mDice and 3.5-5.6% in mIoU across a variety of datasets and scenarios.

Jie Yang, Ziqi Ye, Aihua Ke +3

Key Laboratory of Aerospace Information Security and Trusted Computing, Ministry of Education, School of Cyber Science and Engineering, Wuhan University, Wuhan, China · Shanghai Innovation Institute, Shanghai, China · Shanghai AI Laboratory, Shanghai, China

Synthetic Data Augmentation Medical Image Synthesis Distribution Matching Flow Matching Medical Image Segmentation

Ask PDF

May 6, 2026

FlowDIS: Language-Guided Dichotomous Image Segmentation with Flow Matching

Accurate image segmentation is essential for modern computer vision applications such as image editing, autonomous driving, and medical image analysis. In recent years, Dichotomous Image Segmentation (DIS) has become a standard task for training and evaluating highly accurate segmentation models. Existing DIS approaches often fail to preserve fine-grained details or fully capture the semantic structure of the foreground. To address these challenges, we present FlowDIS, a novel dichotomous image segmentation method built on the flow matching framework, which learns a time-dependent vector field to transport the image distribution to the corresponding mask distribution, optionally conditioned on a text prompt. Moreover, with our Position-Aware Instance Pairing (PAIP) training strategy, FlowDIS offers strong controllability through text prompts, enabling precise, pixel-level object segmentation. Extensive experiments demonstrate that our method significantly outperforms state-of-the-art approaches both with and without language guidance. Compared with the best prior DIS method, FlowDIS achieves a 5.5% higher

F_β^ω

measure and 43% lower MAE (

\mathcal{M}

) on the DIS-TE test set. The code is available at: https://github.com/Picsart-AI-Research/FlowDIS

Andranik Sargsyan, Shant Navasardyan

Picsart AI Research (PAIR)

Image Segmentation Text-Guided Image Segmentation Promptable Image Segmentation Semantic Segmentation

Ask PDF

May 14, 2026

SpectraFlow: Unifying Structural Pretraining and Frequency Adaptation for Medical Image Segmentation

Medical image segmentation remains challenging in low-data regimes, where scarce annotations often yield poor generalization and ambiguous boundaries with missing fine structures. Recent self-supervised pretraining has improved transferability, but it often exhibits a texture bias. In contrast, accurate segmentation is inherently geometry-aware and depends on both topological consistency and precise boundary preservation. To address this problem, we propose a two-stage framework that couples structure-aware encoder pretraining with boundary-oriented decoding. In Stage-1, we aim to learn structure-aware representations for downstream segmentation in low-data regimes. To this end, we propose Mixed-Domain MeanFlow Pretraining, which aligns images and binary masks in a shared latent space through latent transport regression, where masks act as conditional structural guidance rather than prediction targets, making the pretraining task-agnostic. To further improve training stability under scarce supervision, we incorporate a lightweight Dispersive Loss to prevent representation collapse. In Stage-2, we fine-tune the pretrained encoder with a lightweight decoder that combines Direct Attentional Fusion for adaptive cross-scale gating and Frequency-Directional Dynamic Convolution for high-frequency boundary refinement under appearance variation. Experiments on ISIC-2016, Kvasir-SEG, and GlaS demonstrate consistent gains over state-of-the-art methods, with improved robustness in low-data settings and sharper boundary delineation.

Zhiquan Chen, Haitao Wang, Guowei Zou +1

School of Computer Science and Engineering, Sun Yat-sen University, Guangzhou, China · Guangdong Key Laboratory of Big Data Analysis and Processing, Guangzhou, China

Flow Matching Image Segmentation Self-Supervised Pre-Training Medical Image Segmentation Boundary-Aware Segmentation Medical Image Representation Learning+1

Ask PDF

Abstract

Thin-Structure Segmentation

Explore similar work

Few-Shot Distribution-Aligned Flow Matching for Data Synthesis in Medical Image Segmentation

FlowDIS: Language-Guided Dichotomous Image Segmentation with Flow Matching

SpectraFlow: Unifying Structural Pretraining and Frequency Adaptation for Medical Image Segmentation