Multi-Scale Feature Interaction Network

• The paper introduces a lightweight network that leverages hierarchical multi-scale feature extraction to improve image super-resolution performance in resource-constrained environments.
• The architecture features explicit inter-scale interactions and recurrent residual channel attention blocks to efficiently fuse features across scales.
• Experiments demonstrate that MSFIN variants achieve competitive PSNR and SSIM scores while maintaining a favorable accuracy–compactness balance compared to heavier models.

to=arxiv_search 大发游戏官网code {"^^^^^^^^2query2 OR title:\2^^^^^^^^"Lightweight Image Super-Resolution with Multi-scale Feature Interaction Network\"","max_results":5} to=search_arxiv мәҗбур_code {"^^^^^^^^2query2 The Multi-scale Feature Interaction Network (MSFIN) is a lightweight convolutional network for single-image super-resolution (SISR) that reconstructs high-resolution images from low-resolution inputs while targeting deployment scenarios with limited storage and computing resources. Its defining design combines hierarchical multi-scale feature extraction, explicit inter-scale interaction, and a lightweight recurrent residual channel attention block (RRCAB). In the original formulation, MSFIN is organized into shallow feature extraction, multi-scale feature extraction, and reconstruction, and is trained with an PRESERVED_PLACEHOLDER^^^^2query2^^^^ objective (&&&^^^^2query2^^^^&&&).

^^^^2id:(Wang et al., 2021) OR title:\2^^^^. Design objective and problem setting

MSFIN was introduced against a background in which deep and complex convolutional neural networks had already achieved promising SISR performance, but often at the cost of higher memory consumption. The network therefore addresses a specific engineering target: lightweight SISR with a favorable balance between reconstruction quality and model compactness, especially for mobile devices and related resource-constrained settings (&&&^^^^2query2^^^^&&&).

The central design rationale is twofold. First, MSFIN expands the receptive field and exploits informative features from the low-resolution observed image at various scales. Second, it couples those scales through interactive connections rather than treating them as isolated branches. The architecture further incorporates channel attention, but does so through a lightweight block design intended to avoid the parameter growth associated with heavier attention-based super-resolution models (&&&^^^^2query2^^^^&&&).

This combination places MSFIN within a line of super-resolution architectures that seek performance gains from representational organization rather than from simply increasing depth or width. In the terminology of the paper, lightweight behavior is achieved through group convolutions, shallow network design, and parameter sharing via recurrence (&&&^^^^2query2^^^^&&&).

2. End-to-end architecture and optimization

MSFIN is partitioned into three main parts:

^^^^2id:(Wang et al., 2021) OR title:\2^^^^. Shallow Feature Extraction

1. Multi-scale Feature Extraction
2. Reconstruction

Shallow feature extraction applies a single PRESERVED_PLACEHOLDER_^^^^2id:(Wang et al., 2021) OR title:\2^^^^ convolution to the interpolated low-resolution input to obtain low-level features. The full signal path is expressed as

$F_{SF} = H_{sext}(I_{LR}),$

$$F_{DF} = H_{msfe}(F_{SF}) = H_{dext}(H_{msfim}(F_{SF})),$$

$I_{SR} = H_{re}(F_{DF}) = H_{MSFIN}(I_{LR}).$

Here, $H_{sext}$ is the shallow feature extractor, $H_{msfim}$ is the Multi-Scale Feature Interaction Module, $H_{dext}$ denotes deep feature extraction after multi-scale interaction, and $H_{re}$ maps the deep representation back into the high-resolution image space (&&&^^^^2query2^^^^&&&).

Training uses an $L_1$ reconstruction loss:

PRESERVED_PLACEHOLDER_^^^^2id:(Wang et al., 2021) OR title:\2query2^^^^

This formulation is standard for image restoration, but within MSFIN it serves a specifically lightweight architecture whose complexity is concentrated in feature reuse and inter-scale communication rather than in very deep stacks of conventional convolutions (&&&^^^^2query2^^^^&&&).

3. Multi-scale Feature Interaction Module

The core of MSFIN is the Multi-Scale Feature Interaction Module (MSFIM), which performs hierarchical feature extraction and feature interaction across three levels:

• L^^^^2id:(Wang et al., 2021) OR title:\2^^^^: shallow level
• L2: mid-scale level, obtained by PRESERVED_PLACEHOLDER_^^^^2id:(Wang et al., 2021) OR title:\2id:(Wang et al., 2021) OR title:\2^^^^ downsampling
• L3: deepest level, obtained by PRESERVED_PLACEHOLDER_^^^^2id:(Wang et al., 2021) OR title:\22^^^^ downsampling

Each level contains several RRCABs. L2 and L3 use downsampling and upsampling operations, with L3 additionally employing progressive upsampling and local feature fusion. In particular, L3 uses intermediate concatenation followed by a PRESERVED_PLACEHOLDER_^^^^2id:(Wang et al., 2021) OR title:\2^^^^3 convolution to aggregate features across RRCABs (&&&^^^^2query2^^^^&&&).

The distinctive property of MSFIM is that coarse-scale features are not merely decoded back upward; they are injected into finer scales through explicit interconnections. Outputs from deeper levels are fused into shallower levels by transposed convolution, and element-wise summation plus skip connections reintegrate the resulting representations. The final multi-scale fusion is written as

PRESERVED_PLACEHOLDER_^^^^2id:(Wang et al., 2021) OR title:\2^^^^4

where PRESERVED_PLACEHOLDER_^^^^2id:(Wang et al., 2021) OR title:\2^^^^5 denotes a cascade of RRCABs for further feature aggregation (&&&^^^^2query2^^^^&&&).

The paper gives representative interaction equations. At L^^^^2id:(Wang et al., 2021) OR title:\2^^^^,

PRESERVED_PLACEHOLDER_^^^^2id:(Wang et al., 2021) OR title:\2^^^^6

where PRESERVED_PLACEHOLDER_^^^^2id:(Wang et al., 2021) OR title:\2^^^^7 is a transposed convolution used to inject upsampled L2 features into the L^^^^2id:(Wang et al., 2021) OR title:\2^^^^ stream. At L2,

PRESERVED_PLACEHOLDER_^^^^2id:(Wang et al., 2021) OR title:\2^^^^8

where PRESERVED_PLACEHOLDER_^^^^2id:(Wang et al., 2021) OR title:\2^^^^9 denotes features obtained by downsampling $F_{SF} = H_{sext}(I_{LR}),$^^^^2query2^^^^ (&&&^^^^2query2^^^^&&&).

These equations make clear that MSFIN does not restrict information flow to a unidirectional top-down or bottom-up path. Instead, the network organizes feature exchange between coarse and fine resolutions before the final reconstruction layers. A plausible implication is that the model uses coarse levels to enlarge effective context while retaining the finer localization carried by the shallow branch.

4. Lightweight Recurrent Residual Channel Attention Block

MSFIN’s basic computation unit is the lightweight recurrent residual channel attention block, or RRCAB. Its purpose is to model channel dependencies and increase nonlinearity without incurring substantial parameter growth (&&&^^^^2query2^^^^&&&).

RRCAB combines four design elements:

• Group convolution in place of standard convolution, using 6 groups
• $F_{SF} = H_{sext}(I_{LR}),$^^^^2id:(Wang et al., 2021) OR title:\2^^^^ convolution for feature fusion after grouped processing
• Channel attention with SE-like reasoning
• A recurrent structure that loops the block once to increase depth without extra parameters

The channel attention mechanism is formulated as

$F_{SF} = H_{sext}(I_{LR}),$2

where $F_{SF} = H_{sext}(I_{LR}),$3 is global average pooling, $F_{SF} = H_{sext}(I_{LR}),$4 is sigmoid, and $F_{SF} = H_{sext}(I_{LR}),$5 denotes $F_{SF} = H_{sext}(I_{LR}),$6 convolution (&&&^^^^2query2^^^^&&&).

The recurrent residual path is

$F_{SF} = H_{sext}(I_{LR}),$7

with $F_{SF} = H_{sext}(I_{LR}),$8 and $F_{SF} = H_{sext}(I_{LR}),$9 denoting group convolutions and $$F_{DF} = H_{msfe}(F_{SF}) = H_{dext}(H_{msfim}(F_{SF})),$$^^^^2query2^^^^ denoting the recurrence (&&&^^^^2query2^^^^&&&).

Functionally, RRCAB provides a low-overhead way to retain the representational advantages of channel attention in a compact network. The paper’s ablation discussion reports that removing channel attention or feature fusion from RRCAB leads to consistent PSNR reductions, indicating that the block is not merely a parameter-saving device but a core contributor to performance (&&&^^^^2query2^^^^&&&).

5. Model variants, benchmarks, and trade-offs

The paper reports two principal model sizes:

• MSFIN-S: 352K parameters
• MSFIN: 682K parameters

On $$F_{DF} = H_{msfe}(F_{SF}) = H_{dext}(H_{msfim}(F_{SF})),$$^^^^2id:(Wang et al., 2021) OR title:\2^^^^ super-resolution benchmarks including Set5, Set^^^^2id:(Wang et al., 2021) OR title:\2^^^^4, BSDS^^^^2id:(Wang et al., 2021) OR title:\2query2query2^^^^, and Urban^^^^2id:(Wang et al., 2021) OR title:\2query2query2^^^^, the smaller variant is reported to outperform or match existing lightweight models such as CARN-M and s-LWSR32, while the larger variant is comparable to or slightly better than models such as IMDN, CARN, and s-LWSR64 (&&&^^^^2query2^^^^&&&).

Model	Params	Set5 PSNR/SSIM ($$F_{DF} = H_{msfe}(F_{SF}) = H_{dext}(H_{msfim}(F_{SF})),$$2)
CARN-M	4^^^^2id:(Wang et al., 2021) OR title:\22^^^^K	3^^^^2id:(Wang et al., 2021) OR title:\2^^^^.92 / ^^^^2query2^^^^.89^^^^2query2
s-LWSR32	57^^^^2id:(Wang et al., 2021) OR title:\2^^^^K	32.^^^^2query2^^^^4 / ^^^^2query2^^^^.893^^^^2query2
MSFIN-S	352K	32.^^^^2query2^^^^8 / ^^^^2query2^^^^.8934
MSFIN-S+	352K	32.^^^^2id:(Wang et al., 2021) OR title:\2^^^^8 / ^^^^2query2^^^^.8945
MSFIN	682K	32.28 / ^^^^2query2^^^^.8957

The comparison is not limited to Set5. On Urban^^^^2id:(Wang et al., 2021) OR title:\2query2query2^^^^, for example, MSFIN-S+ reaches 25.97dB / ^^^^2query2^^^^.78^^^^2query2, compared with 25.87dB / ^^^^2query2^^^^.779^^^^2query2 for s-LWSR32. For the larger model, the paper reports 32.28dB / ^^^^2query2^^^^.8957 on Set5 for MSFIN, compared with 32.2^^^^2id:(Wang et al., 2021) OR title:\2^^^^dB / ^^^^2query2^^^^.8948 for IMDN. Self-ensemble provides a further increment, with MSFIN+ reaching 32.39dB / ^^^^2query2^^^^.897^^^^2id:(Wang et al., 2021) OR title:\2^^^^ on Set5 (&&&^^^^2query2^^^^&&&).

A particularly concrete trade-off is reported for feature interaction inside MSFIM: introducing feature interaction yields +^^^^2query2^^^^.^^^^2query2 PSNR for +^^^^2id:(Wang et al., 2021) OR title:\2^^^^6K parameters. The ablation study also states that removing channel attention or feature fusion from RRCAB causes consistent PSNR drops. Qualitative comparisons in Fig. 4 are reported to show finer structure recovery and sharper details relative to other lightweight models (&&&^^^^2query2^^^^&&&).

Taken together, these results define MSFIN less as an attempt to maximize absolute reconstruction scores than as an attempt to optimize the accuracy–compactness frontier for lightweight SISR.

6. Position within multi-scale interaction research

Within super-resolution, MSFIN is an explicit formulation of the idea that multi-scale representations should interact rather than merely coexist. That view later appeared in other domains under different architectures and terminology. For example, FIPGNet for salient object detection argues that current methods often “simply aggregate different scale features, ignoring the correlation between different scale features,” and proposes attention-based feature interaction strategies to address that limitation (&&&^^^^2id:(Wang et al., 2021) OR title:\2^^^^8&&&). Similarly, the multistep feature aggregation framework for salient object detection critiques prior “one-way” fusion schemes and introduces multiscale interaction as a separate module (&&&^^^^2id:(Wang et al., 2021) OR title:\2^^^^9&&&).

This suggests that the 2^^^^2query22id:(Wang et al., 2021) OR title:\2^^^^ MSFIN belongs to a broader research movement in which multi-scale processing is increasingly treated as an interaction problem rather than only as a pyramid-construction problem. In that sense, MSFIN’s use of inter-level connections, recurrent lightweight attention blocks, and final aggregation over $$F_{DF} = H_{msfe}(F_{SF}) = H_{dext}(H_{msfim}(F_{SF})),$$3, $$F_{DF} = H_{msfe}(F_{SF}) = H_{dext}(H_{msfim}(F_{SF})),$$4, and $$F_{DF} = H_{msfe}(F_{SF}) = H_{dext}(H_{msfim}(F_{SF})),$$5 anticipates later cross-scale designs in other vision tasks (&&&^^^^2query2^^^^&&&).

A separate source of confusion is nomenclature. The acronym MsFIN was later reused for a traffic accident anticipation model that aggregates short-term, mid-term, and long-term temporal features from dashcam videos (&&&2^^^^2id:(Wang et al., 2021) OR title:\2^^^^&&&). That later usage refers to a different architecture and task. In the SISR literature, however, MSFIN specifically denotes the lightweight image super-resolution network introduced in 2^^^^2query22id:(Wang et al., 2021) OR title:\2^^^^, built around MSFIM and RRCAB (&&&^^^^2query2^^^^&&&).