We establish a new large-scale benchmark that contains 30 ground-truth images and 900 synthetic underwater images of the same scene, called synthetic underwater image dataset (SUID). The proposed SUID creates possibility for a full-reference evaluation of existing technologies for underwater image enhancement and restoration.

Instructions: 

We have released our synthetic underwater image dataset (SUID). Have fun! The SUID is for non-commercial use only. Please enjoy the 900 synthetic underwater images. If you use this dataset, please cite the related paper [G. Hou, X. Zhao, Z. Pan, H. Yang, L. Tan and J. Li, "Benchmarking Underwater Image Enhancement and Restoration, and Beyond," IEEE Access, vol. 8, pp. 122078-122091, 2020.]. Thanks.

Categories:
1287 Views

The dataset comprises of image file s of size 20 x 20 pixels for various types of metals and non-metal.The data collected has been augmented, scaled and modified to represent a number a training set dataset.It can be used to detect and identify object type based on material type in the image.In this process both training data set and test data set can be generated from these image files. 

Instructions: 

## Instruction

The dataset is contained in a zip file named as object_type_material_type.zip.Download it and unzip it.

# command unzip object_type_material_type.zip in linux

# Simply unzip in windows

The folder contains five classes as followed.

 

1.copper 2. iron 3. nickel 4. plastic 5. silver.

 

These are stored as sub-directories under main directory(object_type_material_type).Each sub-directory contains 100 image files in jpg format of size 20 x 20 pixels.

 

Out of these classes 4 are metals type as copper, iron, nickel ,silver and one non-metal type as plastic.These image files can be used as training data set and test dataset as well.

 

Categories:
477 Views

The Dataset

We introduce a novel large-scale dataset for semi-supervised semantic segmentation in Earth Observation: the MiniFrance suite.

Instructions: 

##################################################

The MiniFrance Suite

##################################################

Authors:

Javiera Castillo Navarro, javiera.castillo_navarro@onera.fr

Bertrand Le Saux, bls@ieee.org

Alexandre Boulch, alexandre.boulch@valeo.com

Nicolas Audebert, nicolas.audebert@cnam.fr

Sébastien Lefèvre, sebastien.lefevre@irisa.fr

##################################################

About:

This dataset contains very high resolution RGB aerial images over 16 cities and their surroundings from different regions in France, obtained from IGN's BD ORTHO database (images from 2012 to 2014). Pixel-level land use and land cover annotations are provided, generated by rasterizing Urban Atlas 2012.

##################################################

This dataset is partitioned in three parts, defined by conurbations:

1. Labeled training data: data over Nice and Nantes/Saint Nazaire.

2. Unlabeled training data: data over Le Mans, Brest, Lorient, Caen, Calais/Dunkerque and Saint-Brieuc.

3. Test data: data over Marseille/Martigues, Rennes, Angers, Quimper, Vannes, Clermont-Ferrand, Cherbourg, Lille.

Due to the large-scale nature of the dataset, it is divided in several files to download:

- Images for the labeled training partition: contains RGB aerial images for french departments in the labeled training partition.

- Images for the unlabeled training partition (parts 1, 2 and 3): contain RGB aerial images for french departments in the unlabeled training partition.

- Images for the test partition (parts 1, 2, 3 and 4): contain RGB aerial images for french departments in the partition reserved for evaluation.

- Labels for the labeled partition

- Lists of files by conurbation and partition: contain .txt files that list all images included by city.

Land use maps are available for all images in the labeled training partition of the dataset. We consider here Urban Atlas classes at the second hierarchical level. Available classes are:

- 0: No information

- 1: Urban fabric

- 2: Industrial, commercial, public, military, private and transport units

- 3: Mine, dump and contruction sites

- 4: Artificial non-agricultural vegetated areas

- 5: Arable land (annual crops)

- 6: Permanent crops

- 7: Pastures

- 8: Complex and mixed cultivation patterns

- 9: Orchards at the fringe of urban classes

- 10: Forests

- 11: Herbaceous vegetation associations

- 12: Open spaces with little or no vegetation

- 13: Wetlands

- 14: Water

- 15: Clouds and shadows

##################################################

Citation: If you use this dataset for your work, please use the following citation:

@article{castillo2020minifrance,
title={{Semi-Supervised Semantic Segmentation in Earth Observation: The MiniFrance Suite, Dataset Analysis and Multi-task Network Study}},
author={Castillo-Navarro, Javiera and Audebert, Nicolas and Boulch, Alexandre and {Le Saux}, Bertrand and Lef{\`e}vre, S{\'e}bastien},
journal={Under review.},
year={2020}
}

##################################################

Copyright:

The images in this dataset are released under IGN's "licence ouverte". More information can be found at http://www.ign.fr/institut/activites/lign-lopen-data

The maps used to generate the labels in this dataset come from the Copernicus program, and as such are subject to the terms described here: https://land.copernicus.eu/local/urban-atlas/urban-atlas-2012?tab=metadata

Categories:
501 Views

This dataset was created from all Landsat-8 images from South America in the year 2018. More than 31 thousand images were processed (15 TB of data), and approximately on half of them active fire pixels were found. The Landsat-8 sensor has 30 meters of spatial resolution (1 panchromatic band of 15m), 16 bits of radiometric resolution and 16 days of temporal resolution (revisit). The images in our dataset are in TIFF (geotiff) format with 10 bands (excluding the 15m panchromatic band).

Instructions: 

The images in our dataset are in georeferenced TIFF (geotiff) format with 10 bands. We cropped the original Landsat-8 scenes (with ~7,600 x 7,600 pixels) into image patches with 128 x 128 pixels by using a stride overlap of 64 pixels (vertical and horizontal). The masks are in binary format where True (1) represents fire and False (0) represents background and they were generated from the conditions set by Schroeder et al. (2016). We used the Schroeder conditions to process each patch, producing over 1 million patches with at least one fire pixel and the same amount of patches with no fire pixels, randomly selected from the original images.

The dataset is organized as follow. 

It is divided into South American regions for easy downloading. For each region of South America we have a zip file for images of active fire, its masks, and non-fire images. For example:

 - Uruguay-fire.zip

 - Uruguay-mask.zip

 - Uruguay-nonfire.zip

Within each South American region zip files there are the corresponding zip files to each Landsat-8 WRS (Worldwide Reference System). For example:

- Uruguay-fire.zip;

      - 222083.zip

      - 222084.zip

      - 223082.zip

      - 223083.zip

      - 223084.zip

      - 224082.zip

      - 224083.zip

      - 224084.zip

      - 225081.zip

      - 225082.zip

      - 225083.zip

      - 225084.zip

Within each of these Landsat-8 WRS zip files there are all the corresponding 128x128 image patches for the year 2018. 

 

Categories:
1445 Views

These File is group by 7 different datasets with the task of salient object detect.Each folder is an open data set of SOD, which is composed of multiple JPG files. Each JPG picture corresponds to a annotation picture which exists in PNG format.

Categories:
147 Views

This dataset extends the Urban Semantic 3D (US3D) dataset developed and first released for the 2019 IEEE GRSS Data Fusion Contest (DFC19). We provide additional geographic tiles to supplement the DFC19 training data and also new data for each tile to enable training and validation of models to predict geocentric pose, defined as an object's height above ground and orientation with respect to gravity. We also add to the DFC19 data from Jacksonville, Florida and Omaha, Nebraska with new geographic tiles from Atlanta, Georgia.

Instructions: 

Detailed information about the data content, organization, and file formats is provided in the README files. For image data, individual TAR files for training and validation are provided for each city. Extra training data is also provided in separate TAR files. For point cloud data, individual ZIP files are provided for each city from DFC19. These include the original DFC19 training and validation point clouds with full UTM coordinates to enable experiments requiring geolocation.

Original DFC19 dataset:

https://ieee-dataport.org/open-access/data-fusion-contest-2019-dfc2019

We added new reference data to this extended US3D dataset to enable training and validation of models to predict geocentric pose, defined as an object's height above ground and orientation with respect to gravity. For details, please see our CVPR paper.

CVPR paper on geocentric pose:

http://openaccess.thecvf.com/content_CVPR_2020/papers/Christie_Learning_...

Source Data Attribution

All data used to produce the extended US3D dataset is publicly sourced. Data for DFC19 was derived from public satellite images released for IARPA CORED. New data for Atlanta was derived from public satellite images released for SpaceNet 4, licensed under a Creative Commons Attribution-ShareAlike 4.0 International License. All other commercial satellite images were provided courtesy of DigitalGlobe. U. S. Cities LiDAR and vector data were made publicly available by the Homeland Security Infrastructure Program. 

CORE3D source data: https://spacenetchallenge.github.io/datasets/Core_3D_summary.html

SpaceNet 4 source data: https://spacenetchallenge.github.io/datasets/spacenet-OffNadir-summary.html

Test Sets

Validation data from DFC19 is extended here to include additional data for each tile. Test data is not provided for the DFC19 cities or for Atlanta. Test sets are available for the DFC19 challenge problems on CodaLab leaderboards. We plan to make test sets for all cities available for the geocentric pose problem in the near future. 

Single-view semantic 3D: https://competitions.codalab.org/competitions/20208

Pairwise semantic stereo: https://competitions.codalab.org/competitions/20212

Multi-view semantic stereo: https://competitions.codalab.org/competitions/20216

3D point cloud classification: https://competitions.codalab.org/competitions/20217

References

If you use the extended US3D dataset, please cite the following papers:

G. Christie, K. Foster, S. Hagstrom, G. D. Hager, and M. Z. Brown, "Single View Geocentric Pose in the Wild," Proc. of Computer Vision and Pattern Recognition EarthVision Workshop, 2021.

G. Christie, R. Munoz, K. Foster, S. Hagstrom, G. D. Hager, and M. Z. Brown, "Learning Geocentric Object Pose in Oblique Monocular Images," Proc. of Computer Vision and Pattern Recognition, 2020.

B. Le Saux, N. Yokoya, R. Hansch, and M. Brown, "2019 IEEE GRSS Data Fusion Contest: Large-Scale Semantic 3D Reconstruction [Technical Committees]", IEEE Geoscience and Remote Sensing Magazine, 2019.

M. Bosch, K. Foster, G. Christie, S. Wang, G. D. Hager, and M. Brown, "Semantic Stereo for Incidental Satellite Images," Proc. of Winter Applications of Computer Vision, 2019.

Categories:
1609 Views

In this the  Indian Cautionary Traffic sign data-set has been proposed for classifying the cautionary traffic signs. It is composed of more than 9900 images clustered in 17 different classes. The dataset is trained with different convolutional neural networks and the performance of the classification has been compared and analyzed and achieved high performance when compared with state - of- the - art methodologies

Categories:
2465 Views

Solving the external perception problem for autonomous vehicles and driver-assistance systems requires accurate and robust driving scene perception in both regularly-occurring driving scenarios (termed “common cases”) and rare outlier driving scenarios (termed “edge cases”). In order to develop and evaluate driving scene perception models at scale, and more importantly, covering potential edge cases from the real world, we take advantage of the MIT-AVT Clustered Driving Scene Dataset and build a subset for the semantic scene segmentation task.

Instructions: 

 

MIT DriveSeg (Semi-auto) Dataset is a set of forward facing frame-by-frame pixel level semantic labeled dataset (coarsely annotated through a novel semiautomatic annotation approach) captured from moving vehicles driving in a range of real world scenarios drawn from MIT Advanced Vehicle Technology (AVT) Consortium data.

 

Technical Summary

Video data - Sixty seven 10 second 720P (1280x720) 30 fps videos (20,100 frames)

Class definitions (12) - vehicle, pedestrian, road, sidewalk, bicycle, motorcycle, building, terrain (horizontal vegetation), vegetation (vertical vegetation), pole, traffic light, and traffic sign

 

Technical Specifications, Open Source Licensing and Citation Information

Ding, L., Glazer, M., Terwilliger, J., Reimer, B. & Fridman, L. (2020). MIT DriveSeg (Semi-auto) Dataset: Large-scale Semi-automated Annotation of Semantic Driving Scenes. Massachusetts Institute of Technology AgeLab Technical Report 2020-2, Cambridge, MA. (pdf)

Ding, L., Terwilliger, J., Sherony, R., Reimer, B. & Fridman, L. (2020). MIT DriveSeg (Manual) Dataset. IEEE Dataport. DOI: 10.21227/nb3n-kk46.

 

Attribution and Contact Information

This work was done in collaboration with the Toyota Collaborative Safety Research Center (CSRC). For more information, click here.

For any questions related to this dataset or requests to remove identifying information please contact driveseg@mit.edu.

 

Categories:
1748 Views

Semantic scene segmentation has primarily been addressed by forming representations of single images both with supervised and unsupervised methods. The problem of semantic segmentation in dynamic scenes has begun to recently receive attention with video object segmentation approaches. What is not known is how much extra information the temporal dynamics of the visual scene carries that is complimentary to the information available in the individual frames of the video.

Instructions: 

 

MIT DriveSeg (Manual) Dataset is a forward facing frame-by-frame pixel level semantic labeled dataset captured from a moving vehicle during continuous daylight driving through a crowded city street.

The dataset can be downloaded from the IEEE DataPort or demoed as a video.

 

Technical Summary

Video data - 2 minutes 47 seconds (5,000 frame) 1080P (1920x1080) 30 fps

Class definitions (12) - vehicle, pedestrian, road, sidewalk, bicycle, motorcycle, building, terrain (horizontal vegetation), vegetation (vertical vegetation), pole, traffic light, and traffic sign

 

Technical Specifications, Open Source Licensing and Citation Information

Ding, L., Terwilliger, J., Sherony, R., Reimer, B. & Fridman, L. (2020). MIT DriveSeg (Manual) Dataset for Dynamic Driving Scene Segmentation. Massachusetts Institute of Technology AgeLab Technical Report 2020-1, Cambridge, MA. (pdf)

Ding, L., Terwilliger, J., Sherony, R., Reimer, B. & Fridman, L. (2020). MIT DriveSeg (Manual) Dataset. IEEE Dataport. DOI: 10.21227/mmke-dv03.

 

Related Research

Ding, L., Terwilliger, J., Sherony, R., Reimer. B. & Fridman, L. (2019). Value of Temporal Dynamics Information in Driving Scene Segmentation. arXiv preprint arXiv:1904.00758. (link)

 

Attribution and Contact Information

This work was done in collaboration with the Toyota Collaborative Safety Research Center (CSRC). For more information, click here.

For any questions related to this dataset or requests to remove Identifying information please contact driveseg@mit.edu.

 

Categories:
3340 Views

Synthetic Aperture Radar (SAR) images can be extensively informative owing to their resolution and availability. However, the removal of speckle-noise from these requires several pre-processing steps. In recent years, deep learning-based techniques have brought significant improvement in the domain of denoising and image restoration. However, further research has been hampered by the lack of availability of data suitable for training deep neural network-based systems. With this paper, we propose a standard synthetic data set for the training of speckle reduction algorithms.

Instructions: 

In Virtual SAR we have infused images with varying level of noise, which helps in improving the accuray fo blind denoising task. The holdout set can be created using images from USC SIPI Aerials database and the the provided matlab script (preprocess_holdout.m) tested on Matlab R2019b.

 

The usage for research purposes is for free. If you use this dataset, please cite the following paper along with the dataset: Virtual SAR: A Synthetic Dataset for Deep Learning based Speckle Noise Reduction Algorithms

Categories:
1180 Views

Pages