Update README.md

README.md

@@ -23,17 +23,86 @@ configs:
   - split: test
     path: data/test-*
 ---
-Dataset Status & Upload Plan
+🌾 MIST: Multi-Domain Synthetic Dataset for Rural Driving
+
+MIST is a large-scale multi-domain synthetic dataset designed to address the strong urban-centric bias present in existing autonomous driving benchmarks. While most public driving datasets focus on urban environments captured under daytime and clear weather conditions, MIST targets rural driving scenarios, where visual appearance is primarily shaped by environmental factors rather than dense foreground objects.
+
+Rural environments typically feature sparse traffic, long-range road structures, and background-dominant scenes. In such settings, variations in season, time of day, and weather introduce significant domain shifts that are insufficiently represented in existing datasets.
+
+🎯 Dataset Motivation
+
+The lack of systematically structured rural driving data limits the study of domain generalization and robustness to environmental changes. MIST is designed to enable controlled and balanced analysis of environmental domain shifts by explicitly defining and combining key environmental factors that strongly affect rural driving scenes.
+
+🧩 Structured Multi-Domain Design
+
+MIST defines three independent domain axes and combines them to form 32 distinct domain configurations:
+
+Season (4): Spring, Summer, Autumn, Winter
+
+Time of Day (4): Dawn, Daytime, Dusk, Night
+
+Weather (2): Clear, Overcast
+
+Each domain configuration contains an equal number of samples, ensuring a balanced dataset and preventing unintended domain bias. This structure enables fair comparison across domains and supports systematic evaluation of environmental variations.
+
+🛠️ Dataset Generation
+
+The MIST dataset was generated using Slowroads (https://slowroads.io
+), an open-source procedural driving simulator designed for continuous rural road generation.
+Slowroads enables the creation of diverse rural terrains and long-range road structures through procedural synthesis, making it well-suited for modeling rural driving environments.
+
+All images were rendered from a bumper-view camera to closely capture road surface conditions and environmental appearance during driving. The final image resolution is 1365 × 911, preserving fine-grained visual details of rural scenes.
+
+📦 Dataset Scale & Contents
+
+Total images: ~200,000 synthetic rural driving images
+
+Image resolution: 1365 × 911
+
+Camera viewpoint: Bumper-view
+
+Simulation environment: Slowroads (procedural rural road generation)
+
+Each image is accompanied by:
+
+Instance segmentation masks
+Pixel-level instance annotations for key scene elements such as roads, vegetation, and trees, extracted directly from the simulation engine.
+
+Image–text pairs
+Structured text descriptions encoding the corresponding domain attributes
+(e.g., “winter night overcast rural road”).
+
+This design supports both vision-only and vision–language research, as well as generative modeling tasks.
+
+🔬 Supported Research Tasks
+
+MIST is suitable for a wide range of downstream tasks, including:
+
+Zero-shot domain classification
+
+Multi-domain image-to-image translation
+
+
+
+By providing explicitly structured and balanced rural driving domains, MIST serves as a practical benchmark for studying environmental domain variation beyond urban settings.
+
+📌 Dataset Status & Upload Plan
 
 This repository is being uploaded incrementally.
 
-Current upload:
-We have first uploaded the image-text pair (train/test)dataset intended for image-to-image (I2I) translation tasks.
+✅ Current Upload
+
+Image–text pair dataset (train / test splits)
+
+Intended for image-to-image (I2I) translation and vision–language tasks
+
+🔄 In Progress
+
+Corresponding instance segmentation masks are currently being uploaded
 
-In progress:
-The corresponding segmentation masks are currently being uploaded.
+📅 Upcoming
 
-Upcoming:
-The full dataset, including all remaining data splits and annotations, will be uploaded in subsequent updates.
+The full dataset, including all remaining data splits and annotations, will be released in subsequent updates
 
-Please note that the dataset is not yet complete, and this README will be updated as additional components become available.
+Please note that the dataset is not yet complete.
+This README will be updated as additional components become available.