Article Directory
foreword
The data set corresponds to the application scenario. Different application scenarios have different detection difficulties and corresponding improvement methods. This series organizes and summarizes the data sets in the field, so that everyone can download the data set. Pay attention to receive the organized data set information for free!
1. NWPU VHR-10 Satellite Imagery Dataset
NWPU VHR-10 Dataset is a 10-level geographic remote sensing dataset for space object detection. It has 650 images containing targets and 150 background images, a total of 800. Target types include aircraft, ships, oil tanks, baseballs There are 10 categories including courts, tennis courts, basketball courts, track and field, ports, bridges and cars. This dataset was released by Northwestern Polytechnical University in 2014. Related papers include "Multi-class geospatial object detection and geographic image classification based on collection of part detectors", "A survey on object detection in optical remote sensing images" and "Learningrotation-invariant convolutional Neural networks for object detection in VHRoptical remote sensing images "
Converted format data set: After paying attention, send a private message to receive
the download link【Click here to download】
2. RSOD
RSOD is an open object detection dataset for object detection in remote sensing images. The dataset contains aircraft, fuel tanks, sports fields and overpasses, annotated in the format of the PASCAL VOC dataset.
The dataset consists of 4 folders, each containing one type of object:
1. Aircraft dataset, 4993 aircrafts in 446 images
2. Playgrounds, 191 playgrounds in 189 images.
3. Overpasses, 180 overpasses in 176 images.
4. Tanks, 1586 tanks in 165 images.
Download link【Click here to download】
3. DIOR
"DIOR" is a large-scale benchmark dataset for object detection in optical remote sensing images. The dataset contains 23463 images and 192472 instances, covering 20 object classes. The 20 object classes are airplanes, airports, baseball fields, basketball courts, bridges, chimneys, dams, highway service areas, highway tollbooths, ports, golf courses, ground athletic fields, overpasses, ships, stadiums, storage tanks, Tennis courts, train station, vehicles and wind mill.
Download link【Click here to download】
Convert the format data set: After paying attention, send a private message to receive it
4. DOTA
DOTA is a large-scale dataset for object detection in aerial imagery. It can be used to develop and evaluate object detection in aerial imagery. For the DOTA dataset, it contains 2806 aerial images from different sensors and platforms. Each image ranges in size from approximately 800×800 to 4000×4000 pixels and contains objects of various scales, orientations and shapes. These DOTA images were classified by experts in aerial image interpretation into 15 common object categories. The fully annotated DOTA image contains 188, 282 instances, each labeled by an arbitrary (8 degrees of freedom) quadrilateral.
Download link【Click here to download】
5. HRSC2016
All images in the dataset are from six famous ports. The image resolution is between 2-m and 0.4-m. Image sizes range from 300 to 1500, with most images larger than 1000 x 600. The training, validation and test sets contain 436 images (including 1207 samples), 181 images (including 541 samples) and 444 images (including 1228 samples), respectively.
Download link【Click here to download】
6. UCAS AOD
The UCAS AOD dataset (Zhu et al., 2015a) is used for aircraft and vehicle detection. Specifically, the airplane dataset includes 600 images and 3210 airplanes, while the vehicle dataset includes 310 images and 2819 vehicles. All images are carefully selected so that the object orientations in the dataset are evenly distributed.
7. HRRSD
The HRRSD dataset is a dataset released by the University of Chinese Academy of Sciences in 2019. HRRSD contains 21,761 images obtained from Google Earth and Baidu maps, with spatial resolutions ranging from 0.15-m to 1.2-m. There are 55740 instances of objects in HRRSD, about 4k per class. HRRSD contains 13 categories of targets. The 13 categories are: Airplane, Baseball Field, Basketball Field, Bridge, Intersection, Track Field, Port, Parking Lot, Boat, Storage Tank, T-junction, Tennis Court, Car. The bright spot of the database is that the sample size is relatively balanced among the various categories, and each category has about 4000 samples.
Download link【Click here to download】
Convert the format data set: After paying attention, send a private message to receive it
8. SSDD
In the data set SSDD, there are a total of 1160 images and 2456 ships, with an average of 2.12 ships per image, and the data set will continue to expand. Compared with the PASCAL VOC dataset with 20 types of targets, SSDD has fewer pictures, but the category is only ships, so it is sufficient to train the detection model.
九、DLR 3K Vehicle
DLR 3K Vehicles: The DLR 3K Vehicle dataset (Liu and Mattyus, 2015) is another dataset designed for vehicle detection. It contains 20 5616 3744 aerial images with a spatial resolution of 13 cm. They were filmed at an altitude of 1,000 meters above the ground in the Munich area of Germany using the DLR 3K camera system, a near real-time airborne digital surveillance system. There are 14,235 vehicles manually labeled by using oriented bounding boxes in the images.
To be continued. . .
At present, I have downloaded and compiled the following data sets. Friends who cannot find the download method can pay attention to private messages and receive them.
Reference Link
Remote Sensing Object Detection Dataset