UnderwaterDataset

This repository contains various underwatr datasets for enhancement, detection, segmentation...

Each dataset contains the following types of data:

News.

🚩 There is a Chinese review article about underwater datasets.

🚩 I have found a website where there are many small underwater datasets available for download.

Enhancement/Restoration.

Year	Dataset	Type	Pub	Description
2019	SQUID		TPAMI	It includes RAW images, TIF files, camera clibration files, and distance maps. The database contains 57 stereo pairs from four different sites.
2019	U45		None	A pubic underwater test dataset including 45 images suffering from the color casts, low contrast and haze-like effects of underwater degradation.
2019	Sea-thru Sea-thru		CVPR	It includes RAW images (.ARW or .DEF files) and corresponding depth maps (.tif files).
2019	RUIE		TCSVT	It contains two subset datasets, UIQS and UCCS, which contain only raw underwater images to test UIE algorithms for the ability of improving visibility and correcting color casts, respectively.
2019	UIEB		TIP	UIEB includes two subsets: 890 raw underwater images with corresponding high-quality reference images; 60 challenging underwater images.
2020	EUVP		RAL	There are over 12K paired and 8K unpaired instances in EUVP. Training a CycleGAN on unpaired data to learn the domain transformation between the good and poor quality images. Then the good quality images are distorted by CycleGAN to generate paired images.
2020	OceanDark		CVPRW	It is composed by 183 underwater images of 1280 x 720 pixels captured by video cameras located in profound depths using artificial lighting.
2022	LNRUD		CVPRW	It contains 50000 clean images and 50000 corresponding underwater images synthesized from 5000 real underwater scene images and UWNR (Underwater Neural Rendering).
2023	LSUI		TIP	A large-scale underwater image dataset including 4279 pairs of raw underwater images and corresponding high-quality reference images.
2023	UVE-38K		MTA	The UVE-38K is a large real-world underwater video enhancement dataset with inter-frame consistent refference. It contains 50 video sequences with more than 38,000 frames. They adopt 12 enhancement methods to generate the enhancement candidate pool for each frame and invited volunteers to pick out the best enhancement approach for each video according to their overall performance on the whole video. The videos whose intermediate references include obvious style difference will be applied for further post-refinement to get more consistent and better reference.
2023	DRUVA		ICCV	It contains 20 videos of 20 different artifacts (artifacts are mainly submerged rocks of dimension 0.5 meter to 1 meter) in shallow waters. The videos are approximately 1 minute long and contain an almost 360◦ azimuthal view of the artifacts at a depth of 0.5 meter to 4 meters from the camera.
2023	NUID		TCSVT	It contains 925 real underwater images of different non-uniform illumination.
2024	SUVE		Arxiv	It comprises 660 pairs of training videos and 180 pairs of testing videos, covering a wide range of underwater imaging conditions. Each video contains an average of around 170 frames, resulting in over 140,000 underwater images in the entire dataset.
2024	UVEB		CVPR	It contains 1,308 pairs of video sequences and more than 453,000 high-resolution with 38% Ultra-High-Definition (UHD) 4K frame pairs. UVEB comes from multiple countries, containing various scenes and video degradation types to adapt to diverse and complex underwater environments.

Object Detection.

Year	Dataset	Tag	Pub	Description
2015	FishCLEF-2015		None	It includes 93 videos, manually-annotated with fish locations (as bounding boxes) and species. In total the dataset contains more than 14,000 annotations and 15 classes of fish.
2019	RUIE		TCSVT	It contains a subset dataset, UHTS, which contains 300 images embodying three classes of sea life, i.e., scallop, sea cucumbers and sea urchins.
2019	Brackish		CVPRW	89 videos are provided with annotations in the AAU Bounding Box, YOLO Darknet, and MS COCO formats. Fish are annotated in 6 different coarse categories, i.e., fish, small_fish, crab, shrimp, jellyfish, and starfish.
2019	Trash-ICRA19		ICRA	Some videos were processed to extract 5,700 images, which comprise this dataset, all labeled with bounding boxes on instances of trash, biological objects such as plants and animals, and ROVs.
2021	DUO		ICME	It contains 7782 images, where 6671 for training and 1111 for testing. All the images come from historical URPC datasets and UDD dataset, i.e., echinus, holothurian, starfish and scallop.
2023	FishNet		ICCV	It contains 94,532 meticulously organized images from 17,357 aquatic species, organized according to aquatic biological taxonomy (order, family, genus, and species). All instances in each image are labelled bounding boxes, resulting in 114,375 object instances.

Object Tracking.

Year	Dataset	Tag	Pub	Description
2021	UOT100		IJOE	It consists of 104 video sequences and over 74 000 annotated frames.
2023	UVOT400		Arxiv	It consists of 400 video segments and 275,000 manually annotated frames enabling underwater training and evaluation of deep trackers. The videos are labelled with several underwater-specific tracking attributes including watercolor variation, target distractors, camouflage, target relative size, and low visibility conditions.
2024	WebUOT-1M		Arxiv	It comprises 1.1 million frames across 1,500 video clips filtered from 408 target categories with high-quality bounding boxes for underwater targets. Additionally, it includes language prompts for video sequences, expanding its application areas, e.g., underwater vision-language tracking.

Semantic Segmentation.

Year	Dataset	Tag	Pub	Description
2020	SUIM		IROS	It contains over 1500 images with pixel annotations for eight object categories: fish (vertebrates), reefs (invertebrates), aquatic plants, wrecks/ruins, human divers, robots, and sea-floor.
2020	TrashCan		None	It contains 7212 images and four big classes: trash(any marine debris), ROV(any man-made item), bio(plants and animals) and unknown(unknown objects). Trash objects were additionally tagged by material (e.g., metal, plastic), instance (e.g., cup, bag), along with binary tags indicating overgrowth, significant decay, or crushed/broken items. Bio objects are plants or animals, and in the case of animals, having six classes (crab, fish, eel, shells, starfish, etc). ROV and unknown objects have no additional tags.
2023	UIIS		ICCV	It contains 4,628 images for 7 categories with pixel-level annotations for underwater instance segmentation task: Fish(Fish and other vertebrates), Reefs(Coral reefs and other invertebrates), Aquatic plants( Aquatic plants and flora), Wrecks/ruins( Underwater damaged man-made artifacts), Human divers( Human and the diving equipment they carry), Robots( AUV, ROV and other underwater robots), and Sea-floor(Sea-floor Rocks and reefs on the seafloor).
2024	USIS10K		ICML	It is collected for the task "Salient Instance Segmentation". It contains 10,632 underwater images with pixel-level annotations in 7 categories from various underwater scenes: Fish, Reefs, Plants, Ruins, Human, Robot, and Seafloor.

Classification.

Year	Dataset	Tag	Pub	Description
2012	Fish4Knowledge		ICPR	It consists of 27370 verified fish images. The whole dataset is divided into 23 clusters and each cluster is presented by a representative species, which is based on the synapomorphies characteristic from the extent that the taxon is monophyletic.
2015	FishCLEF-2015		None	It consists of 20,000 sample images and 15 fish categories.
2018	WildFish		ACMMM	It consists of 1000 fish categories with 54,459 unconstrained images.
2023	FishNet		ICCV	It contains 94,532 meticulously organized images from 17,357 aquatic species, organized according to aquatic biological taxonomy (order, family, genus, and species).