https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F7337879%2Fd85902b733a4fecd570ecb10d911b253%2FFigure1.jpg?generation=1712017346000901&alt=media" alt="">

MatSeg Dataset for Zero-Shot Material States Segmentation The dataset contains large-scale synthetic images for training data and highly diverse real-world image benchmarks for testing. Focusing on zero-shot class-agnostic segmentation of materials and their states. This means finding the region of materials states without pre-training on the specific material classes or states. The benchmark contains a wide range of real-world materials and states. For example: wet regions of the surface, scattered dust, minerals of rocks, the sediment of soils, rotten parts of fruits, degraded and corrosive surface regions, food and liquid states, and many others. The focus is on scattered and fragmented materials, as well as soft boundaries partial transition, and partial similarity between regions. It contains both hard segmentation maps and soft and partial similarity annotations for similar but not identical materials. See Readme Files In zips for readers and Technical details.

Synthethic Training Dataset Structure Each folder contains one image and its segmentation map. RGB_RGB.jpg: The image rgb Mask**.png: where ** a number of the mask of a given material, note materials can overlap and values can be between 0-255 (soft). ObjectMaskOcluded.png: Basically the ROI mask means the region that is annotated, anything not marked in this mask is background and is not annotated.

Real-world image Benchmark A benchmark for zero-shot material state segmentation. The benchmark contains 820 real-world images with a wide range of material states and settings. For example: food states (cooked/burned..), plants (infected/dry.) to rocks/soil (minerals/sediment), construction/metals (rusted, worn), liquids (foam/sediment), and many other states in a class-agnostic manner. The goal is to evaluate the segmentation of material materials without knowledge or pretraining on the material or setting. The focus is on materials with complex scattered boundaries, and gradual transition (like the level of wetness of the surface). The annotation of the benchmark is point-based and similarity-based. Hence, for each image, we select several points and regions (Figure 2). We group the points of the same materials into the same label, we also define a group of points that have partial similarity. For example points in group A are more similar to points in group B than to points in group C (In case materials A and B are similar to each other but not identical). This approach allows us to capture the complexity of gradual transition and partial similarities in the world. While also enabling dealing with complex scattered and blurry shapes without needing to annotate the full shape which in many cases is unclear or very hard For more details see : https://arxiv.org/pdf/2403.03309.pdf https://github.com/sagieppel/MatSeg-Dataset For more training data see: https://icedrive.net/s/SBb3g9WzQ5wZuxX9892Z3R4bW8jw

This map shows the seabed data which has been collected in Irish waters. The seabed is mapped using boats and airplanes. The boats use special equipment called a multibeam echosounder. A multibeam echosounder is a type of sonar that is used to map the seabed. Sound waves are emitted in a fan shape beneath the boat. The amount of time it takes for the sound waves to bounce off the bottom of the sea and return to a receiver is used to determine water depth. LiDAR (Light Detection and Ranging) is another way to map the seabed, using airplanes. Two laser light beams are emitted from a sensor on-board an airplane. The red beam reaches the water surface and bounces back; while the green beam penetrates the water hits the seabed and bounces back. The difference in time between the two beams returning allows the water depth to be calculated. LiDAR is only suitable for shallow waters (up to 30m depth).The data are collected as points in XYZ format. X and Y coordinates and Z (depth). The boat travels up and down the water in a series of lines (trackline). An XYZ file is created for each line and contains thousands of points. The line files are merged together and converted into gridded data to create a Digital Terrain Model of the seabed. We use different sized boats and equipment depending on the depth of the water. Some datasets are vector datasets. Vector data portray the world using points, lines, and polygons (areas). Point datasets include the shipwreck locations and seabed sediment sample locations.Line datasets include the sub-bottom profile (rock and sediment below the seabed) tracklines, maritime boundaries and seabed survey tracklines. Polygon datasets include the sub-bottom profile coverage data, INFOMAR/INSS survey zones, priority areas, seabed survey coverage and seabed sediment classification.Some datasets are in raster format. Raster data stores information in a cell-based manner and consists of a matrix of cells (or pixels) organised into rows and columns. The format of the raster is a grid for bathymetry and an image (geotiff) for backscatter. The resolution varies. An example is 10m by 10m cell size for a bathymetry grid. This means that each cell (pixel) represents an area on the seabed of 10 metres squared. Each cell has a depth value which is the average depth of all the points located within that cell. For the backscatter image 10m, it is coloured using grey shades. The darker shading represents a hard seabed (e.g. rock) and lighter shading represents a soft seabed (e.g. sand, silt or mud).This data shows areas that have been surveyed to date. There are plans to fill in the missing areas between 2020 and 2026. The deeper offshore waters were mapped as part of the Irish National Seabed Survey (INSS) between 1999 and 2005. INtegrated Mapping FOr the Sustainable Development of Ireland's MArine Resource (INFOMAR) is mapping the inshore areas. (2006 - 2026).