Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Distilled Pooling Transformer Encoder for Efficient Realistic Image Dehazing

Le-Anh Tran, Dong-Chul Park

TL;DR

Experimental results on various benchmark datasets have shown that the proposed DPTE-Net can achieve competitive dehazing performance when compared to state-of-the-art methods while maintaining low computational complexity, making it a promising solution for resource-limited applications.

Abstract

This paper proposes a lightweight neural network designed for realistic image dehazing, utilizing a Distilled Pooling Transformer Encoder, named DPTE-Net. Recently, while vision transformers (ViTs) have achieved great success in various vision tasks, their self-attention (SA) module's complexity scales quadratically with image resolution, hindering their applicability on resource-constrained devices. To overcome this, the proposed DPTE-Net substitutes traditional SA modules with efficient pooling mechanisms, significantly reducing computational demands while preserving ViTs' learning capabilities. To further enhance semantic feature learning, a distillation-based training process is implemented which transfers rich knowledge from a larger teacher network to DPTE-Net. Additionally, DPTE-Net is trained within a generative adversarial network (GAN) framework, leveraging the strong generalization of GAN in image restoration, and employs a transmission-aware loss function to dynamically adapt to varying haze densities. Experimental results on various benchmark datasets have shown that the proposed DPTE-Net can achieve competitive dehazing performance when compared to state-of-the-art methods while maintaining low computational complexity, making it a promising solution for resource-limited applications. The code of this work is available at https://github.com/tranleanh/dpte-net.

Distilled Pooling Transformer Encoder for Efficient Realistic Image Dehazing

TL;DR

Experimental results on various benchmark datasets have shown that the proposed DPTE-Net can achieve competitive dehazing performance when compared to state-of-the-art methods while maintaining low computational complexity, making it a promising solution for resource-limited applications.

Abstract

This paper proposes a lightweight neural network designed for realistic image dehazing, utilizing a Distilled Pooling Transformer Encoder, named DPTE-Net. Recently, while vision transformers (ViTs) have achieved great success in various vision tasks, their self-attention (SA) module's complexity scales quadratically with image resolution, hindering their applicability on resource-constrained devices. To overcome this, the proposed DPTE-Net substitutes traditional SA modules with efficient pooling mechanisms, significantly reducing computational demands while preserving ViTs' learning capabilities. To further enhance semantic feature learning, a distillation-based training process is implemented which transfers rich knowledge from a larger teacher network to DPTE-Net. Additionally, DPTE-Net is trained within a generative adversarial network (GAN) framework, leveraging the strong generalization of GAN in image restoration, and employs a transmission-aware loss function to dynamically adapt to varying haze densities. Experimental results on various benchmark datasets have shown that the proposed DPTE-Net can achieve competitive dehazing performance when compared to state-of-the-art methods while maintaining low computational complexity, making it a promising solution for resource-limited applications. The code of this work is available at https://github.com/tranleanh/dpte-net.

Paper Structure

This paper contains 25 sections, 10 equations, 17 figures, 8 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Proposed Framework
  4. Architectures
  5. Pooling Transformer Block (PTB)
  6. Framework Architectures
  7. Knowledge Transfer
  8. Two-stage Training
  9. Loss Functions
  10. Experiments and Results
  11. Experimental Setup
  12. Datasets and Competitors
  13. Implementation Settings
  14. Results and Comparisons
  15. Quantitative Evaluation
  16. ...and 10 more sections

Figures (17)

  • Figure 1: Comparison of various methods on the NH-HAZE dataset. The proposed DPTE-Net achieves competitive performance with low computational complexity. The circle radius denotes the model size.
  • Figure 2: The proposed framework.
  • Figure 3: Convergence process of a dehazing network (a) with/(b) without the teacher network's guidance.
  • Figure 4: Two-stage training process.
  • Figure 5: Transmission-aware loss leverages the transmission information in order to optimize the network's generalization based on different degrees of haze density.
  • ...and 12 more figures