anil988/EADF-Net-Paper-Comparison-Images

EADF-Net: Comparative Qualitative Visualizations

This repository contains high-resolution qualitative visual comparisons and benchmark evaluation masks for the research paper:

EADF-Net: Edge-Aware Dual-Stream Feature Fusion Network for Remote Sensing Semantic Segmentation
Evaluated on the official LoveDA (Land-cOVE Domain Adaptation) Benchmark (NeurIPS 2021).

---

📊 Overview Figure (Figure 7 Qualitative Comparison)

Columns shown:

1. (a) Aerial Image: Raw RGB aerial imagery from the LoveDA validation split.
2. (b) Ground Truth: Official benchmark manual human annotations.
3. (c) SegFormer-B5: Vision Transformer benchmark (NeurIPS 2021).
4. (d) D2LS: Dynamic Dictionary Learning (ICCV 2025 SOTA).
5. (e) EADF-Net (Ours): Proposed Edge-Aware Dual-Stream Feature Fusion Network.

---

📦 Direct Download (ZIP Archive)

You can download all 10 evaluation scenes, all model masks, and side-by-side comparison panels in a single ZIP file:

• Download paper_visualizations_loveda.zip (48 MB)

---

📁 Repository Structure

• 01_original_images/: 10 High-resolution LoveDA aerial evaluation scenes (scene_01.png - scene_10.png)
• 02_ground_truth/: Official LoveDA benchmark ground truth segmentation masks
• 03_SegFormer/: Predictions from SegFormer-B5 (IgorNer/segformer-b5-loveda)
• 04_D2LS/: Predictions from D2LS (ICCV 2025 - XavierJiezou/d2ls-models)
• 05_EADF_Net_Ours/: Predictions from EADF-Net (anil988/EADF-Net-LoveDA-CHAMPION)
• 06_composite_panels/: Side-by-side comparison strips and Figure 7 overview (300 DPI)
• paper_visualizations_loveda.zip: All-in-one ZIP archive

---

🎨 LoveDA Color Palette Reference

Class ID	Class Name	Color Representation	RGB Value
0	Background	⬜ White	(255, 255, 255)
1	Building	🟥 Red	(255, 0, 0)
2	Road	🟨 Yellow	(255, 255, 0)
3	Water	🟦 Blue	(0, 0, 255)
4	Barren	🟪 Purple	(159, 129, 183)
5	Forest	🟩 Green	(0, 255, 0)
6	Agriculture	🟧 Light Orange	(255, 195, 128)
255	Void Boundary	⬛ Black	(0, 0, 0)

---

🚀 Quick Download via Python

from huggingface_hub import snapshot_download

# Download all images, masks, panels, and ZIP archive
snapshot_download(repo_id="anil988/EADF-Net-Paper-Comparison-Images", local_dir="./paper_visualizations")