---
title: WaveMamba Fusion Block (WMFB)
url: https://www.emergentmind.com/topics/wavemamba-fusion-block-wmfb
type: topic
---

# WaveMamba Fusion Block (WMFB)

WaveMamba Fusion Block (WMFB) is the core fusion module introduced in “WaveMamba: Wavelet-Driven Mamba Fusion for RGB-Infrared Object Detection” for cross-modality fusion in the wavelet domain. In that formulation, WMFB separates RGB and infrared features into low- and high-frequency sub-bands by Discrete Wavelet Transform (DWT), fuses the low-frequency component with a Mamba-based mechanism, fuses the high-frequency components by an absolute-maximum rule, and supplies the resulting sub-bands to an Inverse Discrete Wavelet Transform (IDWT)-based detection head [2507.18173]. In a broader research context, the label is not standardized: several related papers use different module names for closely related wavelet-guided state-space fusion mechanisms, and some do not use the term “WMFB” at all [2512.17278][2508.11331][2508.03069][2408.01231][2603.02560].

## 1. Terminology and scope

The strict, paper-specific meaning of WMFB comes from WaveMamba for RGB-infrared object detection, where it is explicitly named and described as the module that “facilitates comprehensive fusion across low-/high-frequency sub-bands” [2507.18173]. Outside that paper, the term is used much less consistently. Several later or related works describe functionally similar wavelet-integrated Mamba or state-space fusion units under different names, while one infrared-visible fusion paper provides insufficient extracted technical text to confirm whether “WMFB” appears at all [2503.18378].

| Source | Exact term in source | Relation to WMFB |
|---|---|---|
| [2507.18173] | WMFB | Explicitly introduced |
| [2503.18378] | Wavelet-SSM module; cross-modal feature attention modulation | WMFB-specific details cannot be confirmed |
| [2512.17278] | WHF; DAFF | Closest functional match, but no WMFB term |
| [2508.11331] | HFEB; mask-guided fusion | Treated as WMFB for summary purposes, not explicit |
| [2508.03069] | MDIF | Mapped directly to WMFB in the technical description |
| [2408.01231] | Wavelet Transformation stage | Fusion stage mapped to WMFB |
| [2603.02560] | WSSB; Freq-SSM | Closest practical equivalent |

A common misconception is that WMFB denotes a universally standardized module family. The available papers do not support that reading. The exact name is explicit in the object-detection model of WaveMamba, whereas related segmentation, restoration, and hyperspectral-classification papers describe analogous constructs with task-specific nomenclature and architectural variations [2507.18173][2512.17278][2508.03069].

## 2. Canonical formulation in WaveMamba

In its canonical form, WMFB is embedded in a dual-stream RGB-infrared detector. Features from early backbone stages are decomposed via DWT into four sub-bands per modality: one low-frequency sub-band, $LL$, and three high-frequency sub-bands, $LH$, $HL$, and $HH$ [2507.18173]. The design is driven by a specific empirical observation reported for the M3FD validation set: normalized information entropy is higher in infrared for low-frequency bands and higher in RGB for high-frequency bands. The paper interprets this as indicating that infrared preserves more semantics and shapes at coarse scales, while RGB encodes sharper local edges and textures [2507.18173].

Using Haar wavelets, the default 2D DWT is defined with
$$
L = \frac{1}{\sqrt{2}}[1,1], \qquad H = \frac{1}{\sqrt{2}}[1,-1].
$$
For a feature map $I$, the four sub-bands are
$$
LL = (I \star l_x \star l_y)\downarrow 2,\quad
LH = (I \star l_x \star h_y)\downarrow 2,\quad
HL = (I \star h_x \star l_y)\downarrow 2,\quad
HH = (I \star h_x \star h_y)\downarrow 2,
$$
where $\star$ denotes convolution and $\downarrow 2$ denotes downsampling by 2 in both spatial dimensions [2507.18173]. For an input tensor $f^i$ of shape $B \times C \times H \times W$, each sub-band has shape $B \times C \times (H/2) \times (W/2)$ [2507.18173].

WMFB then decouples fusion into two regimes. Low-frequency sub-bands are fused by the Low-frequency Mamba Fusion Block (LMFB), whereas high-frequency sub-bands are fused independently by a parameter-free high-frequency enhancement rule [2507.18173]. The fused sub-bands are subsequently reconstructed in an improved detection head that replaces conventional upsampling with IDWT, thereby reducing information loss [2507.18173].

Architecturally, WaveMamba applies DWT on the second-stage features $f^2_{RGB}$ and $f^2_{IR}$, then uses multi-level DWT and WMFB on fused features at the 3rd and 5th layers. Ablation identifies three WMFBs placed at stages $\{P2, P3, P5\}$ as the best configuration [2507.18173].

## 3. Low-frequency fusion through LMFB

The low-frequency branch of WMFB is the Low-frequency Mamba Fusion Block, or LMFB. It is built from two stages: Shallow Fusion Mamba (SFM) and Deep Fusion Mamba (DFM) [2507.18173]. SFM performs channel swapping followed by Visual State Space (VSS) blocks, while DFM performs selective state-space fusion with gated attention to suppress redundancy and enhance complementary content [2507.18173].

In channel swapping, the low-frequency RGB and infrared features are split along the channel dimension and partially exchanged. A simple instantiation given in the paper is
$$
F^i_{L,RGB} = [X_A, X_B], \quad F^i_{L,IR} = [Y_A, Y_B],
$$
then
$$
T^i_{L,RGB} = [X_A, Y_B], \quad T^i_{L,IR} = [Y_A, X_B].
$$
The resulting tensors are processed by VSS blocks,
$$
\tilde{F}^i_{L,RGB} = VSS(T^i_{L,RGB}), \qquad
\tilde{F}^i_{L,IR} = VSS(T^i_{L,IR}),
$$
which apply SS2D scanning with selective state updates [2507.18173].

The DFM stage uses two-stream processing per modality. For the primary modality $P$ and auxiliary modality $A$, the paper describes a primary stream with $3\times3$ depth-wise convolution, SiLU, SS2D, and LayerNorm producing $U_P$, an auxiliary stream producing $U_A$, and a gate $G_P$ that regulates both streams. A representative formulation consistent with the paper is
$$
g = \sigma(W_g G_P + b_g), \qquad
Z = W_z[g \odot U_P + (1-g)\odot U_A] + b_z,
$$
optionally with a residual connection [2507.18173]. This is the point at which WMFB most clearly expresses its Mamba-based character: SS2D scans images in four directions to construct sequences tailored to 2D structure, and LayerNorm stabilizes training [2507.18173].

The underlying state-space principle is the generic discrete SSM
$$
h_t = A h_{t-1} + B u_t, \qquad y_t = C h_t + D u_t,
$$
which, in VMamba’s SS2D form, is applied to image-derived sequences rather than to a single rasterized token order [2507.18173]. The intended effect is efficient linear-time modeling of long-range low-frequency semantics without the quadratic cost of self-attention [2507.18173].

## 4. High-frequency fusion and inverse-wavelet reconstruction

WMFB treats high-frequency information differently from low-frequency semantics. Rather than applying another learned Mamba-style fusion stage, WaveMamba fuses each high-frequency sub-band independently by an absolute-maximum rule, denoted HFE in the technical description [2507.18173]. For a high-frequency sub-band $H \in \{LH, HL, HH\}$, the fused output is
$$
F^{hf}_{ij} =
\begin{cases}
H^{rgb}_{ij}, & \text{if } |H^{rgb}_{ij}| \ge |H^{ir}_{ij}|,\\
H^{ir}_{ij}, & \text{if } |H^{ir}_{ij}| > |H^{rgb}_{ij}|.
\end{cases}
$$
Equivalently, the implementation is described by binary masks:
$$
F^i_H = Mask(|F^i_{H,RGB}| - |F^i_{H,IR}|)\odot F^i_{H,RGB}
+ Mask(|F^i_{H,IR}| - |F^i_{H,RGB}|)\odot F^i_{H,IR},
$$
where $Mask(\cdot)$ produces 1 where the argument is positive and 0 otherwise [2507.18173].

This choice reflects a second architectural distinction. A common assumption is that wavelet-Mamba fusion blocks must learn both low- and high-frequency combination. In the original WMFB, that is not the case. Low-frequency fusion is learned and state-space driven; high-frequency fusion is parameter-free and explicitly preserves the strongest edge or detail response at each location [2507.18173].

After fusion, the improved YOLOv8 head reconstructs higher-resolution features by IDWT rather than by nearest or bilinear interpolation. A canonical synthesis formula given in the paper is
$$
\hat{I} = up(LL)\star l_x \star l_y + up(LH)\star l_x \star h_y + up(HL)\star h_x \star l_y + up(HH)\star h_x \star h_y,
$$
where $up(\cdot)$ denotes upsampling by 2 [2507.18173]. The rest of the YOLOv8 head remains unchanged; only the upsampling paths are upgraded to IDWT [2507.18173]. This makes WMFB not merely a fusion unit but part of a larger wavelet-consistent analysis–synthesis pipeline.

## 5. Empirical profile, ablations, and efficiency

WaveMamba reports that it surpasses state-of-the-art approaches with an average mAP improvement of about 4.5% across four benchmarks [2507.18173]. Representative results with the YOLOv8 backbone are reported as follows: on M3FD, $\text{mAP50} = 92.1\%$ and $\text{mAP} = 64.4\%$; on DroneVehicle, $\text{mAP50} = 79.8\%$ and $\text{mAP} = 60.5\%$; on LLVIP, $\text{mAP50} = 98.3\%$ and $\text{mAP} = 66.0\%$; and on FLIR-Aligned, $\text{mAP50} = 88.4\%$ and $\text{mAP} = 48.1\%$ [2507.18173].

The ablation studies isolate the contribution of each WMFB component. On M3FD with the YOLOv8 backbone, removing SFM produces a drop of $1.5\%$ in mAP50 and $2.1\%$ in mAP, while removing DFM produces a drop of $1.9\%$ in mAP50 and $2.2\%$ in mAP [2507.18173]. A fair comparison of fusion strategies with the original head yields the following sequence of results: baseline without DWT, $83.2/55.1$; $(Avg, Avg)$, $86.6/58.2$; $(Avg, LMFB)$, $90.6/62.6$; $(HFE, Avg)$, $90.4/62.3$; and $(HFE, LMFB)$, $91.0/63.3$. Adding the improved head gives the final $(HFE + LMFB + IDWT)$ result of $92.1/64.4$ [2507.18173]. The improved head alone contributes $+1.1\%$ mAP50 and $+1.1\%$ mAP while reducing parameters by $7.6$M [2507.18173].

Wavelet-basis sensitivity is reported to be limited. Performance is robust across MEYR, SYM3, COIF3, and DB3, with Haar achieving the best result of $92.1/64.4$; the differences are reported as $\le 0.4\%$ mAP50 and $\le 0.3\%$ mAP [2507.18173]. With the YOLOv8 backbone, the full model has parameters of approximately $69.1$M and mean inference time of approximately $40$ ms for $640\times640$ pairs on an A800 GPU [2507.18173]. On FLIR-Aligned, the paper states that WaveMamba yields higher accuracy with fewer parameters and faster inference than attention-heavy fusion baselines such as CrossFormer [2507.18173].

These results position WMFB as a module whose empirical contribution depends on the coordinated action of three elements: LMFB for low-frequency semantics, HFE for high-frequency detail preservation, and IDWT-based reconstruction in the head [2507.18173].

## 6. Related formulations, extensions, and naming ambiguity

Several subsequent works adopt closely related design principles while avoiding the exact term WMFB. In “WDFFU-Mamba,” the paper states that it does not use the term “WaveMamba Fusion Block (WMFB)” and instead introduces a Wavelet-denoised High-Frequency-guided Feature (WHF) module and a Dual Attention Feature Fusion (DAFF) module. Functionally, DAFF integrating WHF at the bottleneck is described as the closest match to what one might call a WMFB [2512.17278]. In “Guiding WaveMamba with Frequency Maps for Image Debanding,” the paper likewise does not use the term explicitly; for summary purposes, the High-Frequency Enhance Block (HFEB), together with mask-guided fusion into inverse wavelet reconstruction, is treated as the WMFB-like component [2508.11331].

A similar pattern appears in three-dimensional and hyperspectral settings. “SSFMamba” states that the paper does not explicitly define a WMFB and instead names its fusion unit the Multi-domain Information Fusion Block (MDIF), with a 3D Multi-Directional Scanning Mechanism (MDSM) inside a Mamba block [2508.03069]. “WaveMamba: Spatial-Spectral Wavelet Mamba for Hyperspectral Image Classification” also states that the paper does not explicitly define a module named WMFB; the relevant fusion is implemented by a wavelet transformation stage together with a spatial and spectral gate and concatenation of sub-bands into a fused tensor $E$ that is passed to the Mamba/state-space model [2408.01231]. “CAWM-Mamba” again does not name a module WMFB; the closest fully specified components are the Cross-modal Feature Interaction Module (CFIM), the Wavelet Space State Block (WSSB), and its Frequency-Selective State Space Module (Freq-SSM), with a practical WMFB described as CFIM $\rightarrow$ WSSB $\rightarrow$ weather-guided gating [2603.02560].

The infrared-visible fusion paper “Exploring State Space Model in Wavelet Domain” introduces a “Wavelet-SSM module” and “cross-modal feature attention modulation,” but the provided paper text contains no technical content beyond metadata, so the exact terminology used in the paper, the architecture and hyperparameters of the WaveMamba Fusion Block, loss functions, complexity numbers, or comparison results cannot be confirmed from the extracted material [2503.18378]. This is important because it prevents the term WMFB from being retrospectively attributed to that paper on documentary grounds.

Taken together, these papers suggest a broader conceptual pattern: a “WaveMamba fusion block,” in an editorial rather than authorial sense, typically denotes a module that combines wavelet-domain decomposition with state-space modeling, uses task-specific rules to merge low- and high-frequency information, and often couples that fusion to an IDWT-based reconstruction path [2512.17278][2508.11331][2508.03069][2408.01231][2603.02560]. The exact implementation, however, varies sharply with task, dimensionality, and modality.

## 7. Conceptual significance and common misunderstandings

WMFB is best understood as a frequency-aware fusion operator rather than as a generic Mamba block. Its defining feature in the canonical WaveMamba detector is the explicit asymmetry between low- and high-frequency processing: low-frequency fusion is learned through Mamba-based selective scanning and gated interaction, whereas high-frequency fusion is delegated to an absolute-maximum selector [2507.18173]. This division encodes a domain assumption reported by the paper’s entropy analysis: infrared is more informative in low-frequency bands, while RGB is more informative in high-frequency bands [2507.18173].

A second misunderstanding is that WMFB is synonymous with wavelet-domain fusion in infrared-visible image fusion. The available record does not support that equivalence. Some papers use wavelet-domain state-space modules for fusion, segmentation, enhancement, or restoration, but under names such as Wavelet-SSM, WHF, DAFF, HFEB, MDIF, or WSSB, and with different placement, loss design, and reconstruction strategy [2503.18378][2512.17278][2508.11331][2508.03069][2603.02560]. WMFB is therefore both a specific module name and, only by plausible extension, a broader family resemblance.

A plausible implication is that the enduring contribution of WMFB lies less in a single fixed blueprint than in an architectural doctrine: decompose heterogeneous features into wavelet sub-bands, assign low-frequency semantics and high-frequency detail to different fusion mechanisms, and exploit state-space models where global context is most needed [2507.18173][2512.17278][2603.02560]. Under that reading, the original WaveMamba formulation remains the canonical reference, while later variants show how the same principle can be adapted to detection, segmentation, restoration, and hyperspectral classification.

Source: https://www.emergentmind.com/topics/wavemamba-fusion-block-wmfb