Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SwinIFS: Landmark Guided Swin Transformer For Identity Preserving Face Super Resolution

Habiba Kausar, Saeed Anwar, Omar Jamal Hammad, Abdul Bais

TL;DR

Face super-resolution under severe degradation is challenging due to loss of identity cues. SwinIFS addresses this by fusing dense landmark heatmaps with a compact Swin Transformer backbone and Residual Swin Transformer Blocks to jointly model texture and geometry, guided by explicit structural priors. The pipeline uses landmark-augmented input, hierarchical transformer refinement, PixelShuffle upsampling, and a hybrid $\ell_1$ plus perceptual loss, achieving state-of-the-art results on CelebA at $4\times$ and $8\times$ while maintaining real-time-like efficiency. This approach offers strong identity preservation and perceptual quality, with practical potential for facial enhancement, surveillance, and digital restoration.

Abstract

Face super-resolution aims to recover high-quality facial images from severely degraded low-resolution inputs, but remains challenging due to the loss of fine structural details and identity-specific features. This work introduces SwinIFS, a landmark-guided super-resolution framework that integrates structural priors with hierarchical attention mechanisms to achieve identity-preserving reconstruction at both moderate and extreme upscaling factors. The method incorporates dense Gaussian heatmaps of key facial landmarks into the input representation, enabling the network to focus on semantically important facial regions from the earliest stages of processing. A compact Swin Transformer backbone is employed to capture long-range contextual information while preserving local geometry, allowing the model to restore subtle facial textures and maintain global structural consistency. Extensive experiments on the CelebA benchmark demonstrate that SwinIFS achieves superior perceptual quality, sharper reconstructions, and improved identity retention; it consistently produces more photorealistic results and exhibits strong performance even under 8x magnification, where most methods fail to recover meaningful structure. SwinIFS also provides an advantageous balance between reconstruction accuracy and computational efficiency, making it suitable for real-world applications in facial enhancement, surveillance, and digital restoration. Our code, model weights, and results are available at https://github.com/Habiba123-stack/SwinIFS.

SwinIFS: Landmark Guided Swin Transformer For Identity Preserving Face Super Resolution

TL;DR

Face super-resolution under severe degradation is challenging due to loss of identity cues. SwinIFS addresses this by fusing dense landmark heatmaps with a compact Swin Transformer backbone and Residual Swin Transformer Blocks to jointly model texture and geometry, guided by explicit structural priors. The pipeline uses landmark-augmented input, hierarchical transformer refinement, PixelShuffle upsampling, and a hybrid plus perceptual loss, achieving state-of-the-art results on CelebA at and while maintaining real-time-like efficiency. This approach offers strong identity preservation and perceptual quality, with practical potential for facial enhancement, surveillance, and digital restoration.

Abstract

Face super-resolution aims to recover high-quality facial images from severely degraded low-resolution inputs, but remains challenging due to the loss of fine structural details and identity-specific features. This work introduces SwinIFS, a landmark-guided super-resolution framework that integrates structural priors with hierarchical attention mechanisms to achieve identity-preserving reconstruction at both moderate and extreme upscaling factors. The method incorporates dense Gaussian heatmaps of key facial landmarks into the input representation, enabling the network to focus on semantically important facial regions from the earliest stages of processing. A compact Swin Transformer backbone is employed to capture long-range contextual information while preserving local geometry, allowing the model to restore subtle facial textures and maintain global structural consistency. Extensive experiments on the CelebA benchmark demonstrate that SwinIFS achieves superior perceptual quality, sharper reconstructions, and improved identity retention; it consistently produces more photorealistic results and exhibits strong performance even under 8x magnification, where most methods fail to recover meaningful structure. SwinIFS also provides an advantageous balance between reconstruction accuracy and computational efficiency, making it suitable for real-world applications in facial enhancement, surveillance, and digital restoration. Our code, model weights, and results are available at https://github.com/Habiba123-stack/SwinIFS.
Paper Structure (14 sections, 10 equations, 4 figures, 1 table)

This paper contains 14 sections, 10 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. SwinIFS
  4. Landmark Encoding and Input Construction
  5. Shallow and Deep Feature Extraction
  6. Residual Swin Transformer Block
  7. Reconstruction and Upsampling
  8. Loss Functions
  9. Experiments
  10. Dataset and Preprocessing
  11. Evaluation Metrics
  12. Experimental Implementation Details
  13. Results and Discussion
  14. Conclusion

Figures (4)

  • Figure 1: Overview of the proposed SwinIFS framework. The pipeline begins with landmark-guided input construction, in which the LR image is combined with five Gaussian heatmaps to form an 8-channel tensor. Shallow feature extraction projects this tensor into a high-dimensional embedding, followed by deep hierarchical refinement using stacked RSTBs and STLs. Finally, a reconstruction and PixelShuffle upsampling module synthesizes the high-resolution face image, preserving identity and restoring fine structural details.
  • Figure 2: Visual comparison of face super-resolution results for $4\times$ upscaling on CelebA. SwinIFS produces sharper and more identity-preserving reconstructions than competing methods.
  • Figure 4: Region-specific comparison of eye and mouth reconstruction for $4\times$ upscaling. SwinIFS maintains fine structural cues essential for identity preservation.
  • Figure 6: PSNR versus inference time for $8\times$ face super-resolution models. SwinIFS achieves the best balance between reconstruction quality and computational efficiency.