Update README.md

README.md

@@ -11,18 +11,19 @@ tags:
 - fiftyone
 - image
 - image-segmentation
-dataset_summary: '
+dataset_summary: >
 
 
 
 
-  This is a [FiftyOne](https://github.com/voxel51/fiftyone) dataset with 4420 samples.
+  This is a [FiftyOne](https://github.com/voxel51/fiftyone) dataset with 4420
+  samples.
 
 
   ## Installation
 
 
-  If you haven''t already, install FiftyOne:
+  If you haven't already, install FiftyOne:
 
 
   ```bash
@@ -44,7 +45,7 @@ dataset_summary: '
 
   # Load the dataset
 
-  # Note: other available arguments include ''max_samples'', etc
+  # Note: other available arguments include 'max_samples', etc
 
   dataset = load_from_hub("harpreetsahota/RefSegRS")
 
@@ -54,16 +55,12 @@ dataset_summary: '
   session = fo.launch_app(dataset)
 
   ```
-
-  '
+license: cc-by-4.0
 ---
 
-# Dataset Card for refsegrs
-
-<!-- Provide a quick summary of the dataset. -->
-
-
+# Dataset Card for RefSegRS
 
+![image/png](refseg_rs.gif)
 
 
 This is a [FiftyOne](https://github.com/voxel51/fiftyone) dataset with 4420 samples.
@@ -84,141 +81,179 @@ from fiftyone.utils.huggingface import load_from_hub
 
 # Load the dataset
 # Note: other available arguments include 'max_samples', etc
-dataset = load_from_hub("harpreetsahota/RefSegRS")
+dataset = load_from_hub("Voxel51/RefSegRS")
 
 # Launch the App
 session = fo.launch_app(dataset)
+
+
 ```
+#### NOTE: This dataset is .tif media, for best results view in Safari browser or a browser which supports displaying .tif
 
 
 ## Dataset Details
 
 ### Dataset Description
 
-<!-- Provide a longer summary of what this dataset is. -->
+RefSegRS is a referring remote sensing image segmentation (RRSIS) dataset that enables pixel-level segmentation of objects in remote sensing imagery based on natural language descriptions. The dataset addresses the task of localizing and segmenting desired objects from remote sensing images using referring expressions that include categories, attributes, and spatial relationships.
+
+The dataset is built on top of the SkyScapes dataset, consisting of cropped and downsampled aerial RGB images with corresponding segmentation masks and natural language referring expressions. Images are captured from a top-down view with 13 cm spatial resolution, featuring urban scenes with various objects including vehicles, roads, buildings, vegetation, and infrastructure elements.
 
+- **Curated by:** Zhenghang Yuan, Lichao Mou, Yuansheng Hua, and Xiao Xiang Zhu (Technical University of Munich)
+- **Language(s) (NLP):** English
+- **License:** CC-BY-4.0
 
+### Dataset Sources
 
-- **Curated by:** [More Information Needed]
-- **Funded by [optional]:** [More Information Needed]
-- **Shared by [optional]:** [More Information Needed]
-- **Language(s) (NLP):** en
-- **License:** [More Information Needed]
+- **HF Repository:** https://huggingface.co/datasets/JessicaYuan/RefSegRS
+- **Project Repository:** https://github.com/zhu-xlab/rrsis
+- **Paper (arXiv):** https://arxiv.org/abs/2306.08625
+- **Related Work:** This dataset is part of research on combining remote sensing imagery with natural language processing, related to visual grounding, visual question answering (VQA), and image captioning for remote sensing data.
 
-### Dataset Sources [optional]
 
-<!-- Provide the basic links for the dataset. -->
+## Dataset Structure
 
-- **Repository:** [More Information Needed]
-- **Paper [optional]:** [More Information Needed]
-- **Demo [optional]:** [More Information Needed]
+The RefSegRS dataset contains **4,420 image-language-label triplets** organized into three splits:
 
-## Uses
+- **Training set:** 2,172 triplets
+- **Validation set:** 431 triplets  
+- **Test set:** 1,817 triplets
 
-<!-- Address questions around how the dataset is intended to be used. -->
+### Image Specifications
 
-### Direct Use
+- **Format:** TIFF (RGB, 3 channels)
+- **Dimensions:** 512 × 512 pixels
+- **Original resolution:** 13 cm spatial resolution
+- **Source:** Cropped from SkyScapes dataset tiles (original 5616 × 3744 pixels) using 1200 × 1200 pixel sliding windows with 600-pixel stride, then downsampled
 
-<!-- This section describes suitable use cases for the dataset. -->
+### Segmentation Masks
 
-[More Information Needed]
+- **Format:** TIFF (binary masks)
+- **Dimensions:** 512 × 512 pixels
+- **Values:** Binary (0 for background, 1 for target object)
+- **Generation:** Automatically generated from SkyScapes pixel-wise annotations based on referring expressions
 
-### Out-of-Scope Use
+### Object Categories
 
-<!-- This section addresses misuse, malicious use, and uses that the dataset will not work well for. -->
+The dataset includes 20 object categories from the SkyScapes dataset:
+- **Vegetation:** low vegetation, tree
+- **Roads:** paved road, non-paved road, bikeway, sidewalk, lane marking
+- **Parking:** paved parking place, non-paved parking place
+- **Vehicles:** car, trailer, van, truck, large truck, bus
+- **Infrastructure:** building, entrance/exit, danger area
+- **Other:** clutter, impervious surface
 
-[More Information Needed]
+### Referring Expressions
 
-## Dataset Structure
+Natural language descriptions are generated using templates that include:
+- **Categories:** Direct object names (e.g., "vehicle", "road")
+- **Attributes:** Object properties (e.g., "light-duty vehicle", "heavy-duty vehicle", "long truck")
+- **Spatial relationships:** Positional descriptions (e.g., "vehicle in the parking area", "light-duty vehicle driving on the road", "building with a parking lot")
 
-<!-- This section provides a description of the dataset fields, and additional information about the dataset structure such as criteria used to create the splits, relationships between data points, etc. -->
+Common expressions include: "car", "road", "impervious surface", "road marking", "vehicle in the parking area", "building along the road", "sidewalk along with tree"
 
-[More Information Needed]
+### FiftyOne Dataset Structure
+
+When loaded into FiftyOne, the dataset has the following structure:
+
+**Sample fields:**
+- `filepath`: Absolute path to the image file
+- `tags`: List containing the split name ("train", "test", or "val")
+- `metadata`: Image metadata (dimensions, size, MIME type)
+- `segmentation`: FiftyOne Segmentation object with absolute path to the mask file
+- `phrase`: String containing the natural language referring expression
+- `created_at`: Timestamp of sample creation
+- `last_modified_at`: Timestamp of last modification
 
 ## Dataset Creation
 
 ### Curation Rationale
 
-<!-- Motivation for the creation of this dataset. -->
+The RefSegRS dataset was created to address the lack of referring image segmentation datasets for remote sensing imagery. While referring image segmentation has been extensively studied for natural images, almost no research attention had been given to this task in the context of remote sensing.
 
-[More Information Needed]
+The dataset enables:
+- End users without domain expertise to obtain precise information from remote sensing imagery using natural language
+- Targeted image analysis where users can specify objects of interest based on their individual needs
+- Improved efficiency and user interactivity in remote sensing image interpretation
 
-### Source Data
+The dataset specifically addresses challenges unique to remote sensing imagery:
+- Small and scattered objects (vehicles, road markings) that occupy fewer pixels
+- Wide range of object categories in top-down views
+- Objects with great scale variations
+- Spatial relationships between objects in urban scenes
 
-<!-- This section describes the source data (e.g. news text and headlines, social media posts, translated sentences, ...). -->
+### Source Data
 
 #### Data Collection and Processing
 
-<!-- This section describes the data collection and processing process such as data selection criteria, filtering and normalization methods, tools and libraries used, etc. -->
+**Image Collection:**
+1. Source images from the SkyScapes dataset (16 RGB tiles, each 5616 × 3744 pixels, 13 cm spatial resolution)
+2. Crop tiles into 1200 × 1200 pixel images using sliding window with 600-pixel stride
+3. Downsample to 512 × 512 pixels to match deep neural network input requirements
 
-[More Information Needed]
+**Referring Expression Generation:**
+- Expressions generated using predefined templates based on how end users typically refer to objects
+- Templates include: category alone, category with attributes, and spatial relationships with other entities
+- Manual filtering performed to remove uninformative image-language-label triplets
 
-#### Who are the source data producers?
+**Mask Generation:**
+1. Pixel-wise annotations sourced from SkyScapes dataset (each pixel labeled with one of 20 classes)
+2. Automatic generation of binary ground truth masks based on natural language expressions
+3. Two types of conceptual relationships established:
+   - **Identity:** Direct mapping (e.g., "road marking" ≡ "lane marking")
+   - **Inclusion:** Hierarchical grouping (e.g., "light-duty vehicle" includes "car" and "van")
 
-<!-- This section describes the people or systems who originally created the data. It should also include self-reported demographic or identity information for the source data creators if this information is available. -->
+#### Who are the source data producers?
 
-[More Information Needed]
+The source imagery comes from the **SkyScapes dataset**, which provides aerial RGB imagery with pixel-wise semantic annotations of urban scenes.
 
-### Annotations [optional]
+The RefSegRS dataset was curated by researchers at:
+- **Technical University of Munich** (Chair of Data Science in Earth Observation)
+- **Shenzhen University** (College of Civil and Transportation Engineering)
+- **Munich Center for Machine Learning**
 
-<!-- If the dataset contains annotations which are not part of the initial data collection, use this section to describe them. -->
+### Annotations
 
 #### Annotation process
 
-<!-- This section describes the annotation process such as annotation tools used in the process, the amount of data annotated, annotation guidelines provided to the annotators, interannotator statistics, annotation validation, etc. -->
-
-[More Information Needed]
-
-#### Who are the annotators?
-
-<!-- This section describes the people or systems who created the annotations. -->
-
-[More Information Needed]
-
-#### Personal and Sensitive Information
-
-<!-- State whether the dataset contains data that might be considered personal, sensitive, or private (e.g., data that reveals addresses, uniquely identifiable names or aliases, racial or ethnic origins, sexual orientations, religious beliefs, political opinions, financial or health data, etc.). If efforts were made to anonymize the data, describe the anonymization process. -->
-
-[More Information Needed]
+The annotations in RefSegRS consist of two components:
 
-## Bias, Risks, and Limitations
+**1. Segmentation Masks:**
+- Automatically generated from existing SkyScapes pixel-wise semantic annotations
+- Binary masks created by setting pixels within the target category to 1 and outside to 0
+- For composite categories (e.g., "vehicle"), masks combine multiple sub-categories ("car", "van", "truck", etc.)
 
-<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+**2. Referring Expressions:**
+- Generated using predefined templates that reflect natural user language patterns
+- Templates incorporate:
+  - Category names (direct specification)
+  - Attributes (size, type, material properties)
+  - Spatial relationships (location, proximity to other objects)
+- Manual filtering applied to remove uninformative or ambiguous triplets
+- Final dataset: 4,420 curated image-language-label triplets
 
-[More Information Needed]
+**Quality Control:**
+- Manual review to ensure referring expressions accurately describe the corresponding masks
+- Filtering of uninformative samples to maintain dataset quality
 
-### Recommendations
-
-<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
-
-Users should be made aware of the risks, biases and limitations of the dataset. More information needed for further recommendations.
+#### Who are the annotators?
 
-## Citation [optional]
+- **Segmentation masks:** Derived from the SkyScapes dataset's existing pixel-wise annotations
+- **Referring expressions:** Generated automatically using templates, then manually filtered by the research team at Technical University of Munich
 
-<!-- If there is a paper or blog post introducing the dataset, the APA and Bibtex information for that should go in this section. -->
+## Citation
 
 **BibTeX:**
 
-[More Information Needed]
+```bibtex
+@article{yuan2023rrsis,
+  title={RRSIS: Referring Remote Sensing Image Segmentation},
+  author={Yuan, Zhenghang and Mou, Lichao and Hua, Yuansheng and Zhu, Xiao Xiang},
+  journal={arXiv preprint arXiv:2306.08625},
+  year={2023}
+}
+```
 
 **APA:**
 
-[More Information Needed]
-
-## Glossary [optional]
-
-<!-- If relevant, include terms and calculations in this section that can help readers understand the dataset or dataset card. -->
-
-[More Information Needed]
-
-## More Information [optional]
-
-[More Information Needed]
-
-## Dataset Card Authors [optional]
-
-[More Information Needed]
-
-## Dataset Card Contact
-
-[More Information Needed]
+Yuan, Z., Mou, L., Hua, Y., & Zhu, X. X. (2024). RRSIS: Referring Remote Sensing Image Segmentation. *IEEE Transactions on Geoscience and Remote Sensing*. arXiv:2306.08625v2