DeepFish as a benchmark suite with a large-scale dataset to train and test methods for several computer vision tasks. The dataset consists of approximately 40 thousand images collected underwater from 20 habitats in the marine environments of tropical Australia. It contains classification labels as well as point-level and segmentation labels to have a more comprehensive fish analysis benchmark. These labels enable models to learn to automatically monitor fish count, identify their locations, and estimate their sizes.

Image bank of fish trays collected in the wholesale fish market in El Campello (Alicante, Spain) by artisanal fishing belonging to the DeepFish project.

The original fish tray images are provided in the "fish_tray_images_2021_MM_DD.zip" files. MM and DD stand for the month and initial day (e.g. 04_01 stands for the first of April and subsequent days, and 05_17 stands for the 17th of May, and subsequent days until the end of the month). The last zip file of this kind, 2021_06-09, contains all images from June to September.

JSON files (in fish_tray_json_labels.zip) are prepared to be used with the "Django Labeller" software, but can be converted to any format, e.g. "COCO" which is also provided in the "coco_format_fish_data.json" file.

Each of these JSON files is composed by an object containing the name of the image and the labels appearing in it. Inside each label, the following information is provided:

Type of label. It can be a size (total, diameter of the eye...), tray or fish specie.

Class of the label. It means the concrete specie, measurement or tray depending on the type of label.

Semantic segmentation represented by one or multiple regions in case of occlusions. Represented by an array of coordinates in the image (x and y).

Object_id: Identifier of the label, unique in the entire dataset.

Father_object_id: In case this is not the main object (The label with the segmentation of the species). It will point to the identifier (ID) of that main species to which it belongs. It means, if this is the total size, it will point to the fish sized like that.

Furthermore, estimated fish sizes are also provided in the "size_estimation_homography_DeepFish.csv" file. These size estimations are calculated using homography of the known tray size, to convert from pixel units to centimetres.

Deepfish

## Overview

Deepfish is a dataset for object detection tasks - it contains Fish annotations for 4,505 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [CC BY 4.0 license](https://creativecommons.org/licenses/CC BY 4.0).

Image bank of fish trays collected in the Campello fish market by artisanal fishing belonging to the DeepFish project.

JSON files are prepared to be used with the program "Django Labeller", but it can be converted to any format, f.e. "COCO". Indeed, the file "data_fish.json" contains the previous information with the images filtered in this COCO format.

Each JSON file is composed by an object containing the name of the image and the labels appearing in it. Inside each label we will have:

Type of label. It can be a size (total, diameter of the eye...), tray or fish specie.

Class of the label. It means the concrete specie, measurement or tray depending on the type of label.

Semantic segmentation represented by one or multiple regions in case of occlusions. Represented by an array of coordinates in the image (x and y).

Object_id: Identifier of the label, unique in the entire dataset.

Father_object_id: In case this is not the main object (The label with the segmentation of the specie). It will point to the id of that main specie which it belongs. It means, if this is the total size, it will point to the fish sized like that.

AnnotateDeepFish

## Overview

AnnotateDeepFish is a dataset for object detection tasks - it contains Fish annotations for 1,560 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [CC BY 4.0 license](https://creativecommons.org/licenses/CC BY 4.0).

MAE result on DeepFish dataset.

Dataset

This dataset was created by DucMinhPhy

Contents

A dataset for fish tracking and segmentation in underwater videos

This dataset use for MSc project,

original data from [1]Norfisk and [2]DeepFish

[1] Crescitelli, A.M., Gansel, L.C. and Zhang, H. (2021b) “NorFisk: fish image dataset from Norwegian fish farms for species recognition using deep neural networks,” Modeling Identification and Control, 42(1), pp. 1–16. Available at: https://doi.org/10.4173/mic.2021.1.1.

[2] Guilló A. F., Lopez J. A., D'Urso N. E., and Cuenca A. G., Capdepon G. S., Maestre M. V., Nieto J. E. G., and Sanchez P P.. (2022). DeepFish Dataset (April 2022 update) (v3.1) [Dataset]. Zenodo. https://doi.org/10.5281/zenodo.6475675

The provided datasets are used for the analysis in the work "Transfer learning with generative models for object detection on limited datasets" (https://doi.org/10.1088/2632-2153/ad65b5). The availability of data is limited in some fields, especially for object detection tasks, where it is necessary to have correctly labeled bounding boxes around each object. A notable example of such data scarcity is found in the domain of marine biology, where it is useful to develop methods to automatically detect submarine species for environmental monitoring. To address this data limitation, the state-of-the-art machine learning strategies employ two main approaches. The first involves pretraining models on existing datasets before generalizing to the specific domain of interest. The second strategy is to create synthetic datasets specifically tailored to the target domain using methods like copy-paste techniques or ad-hoc simulators. The first strategy often faces a significant domain shift, while the second demands custom solutions crafted for the specific task. In response to these challenges, here we propose a transfer learning framework that is valid for a generic scenario. In this framework, generated images help to improve the performances of an object detector in a few-real data regime. This is achieved through a diffusion-based generative model that was pretrained on large generic datasets. With respect to the state-of-the-art, we find that it is not necessary to fine tune the generative model on the specific domain of interest. We believe that this is an important advance because it mitigates the labor-intensive task of manual labeling the images in object detection tasks. We validate our approach focusing on fishes in an underwater environment, and on the more common domain of cars in an urban setting. Our method achieves detection performance comparable to models trained on thousands of images, using only a few hundreds of input data. Our results pave the way for new generative AI-based protocols for machine learning applications in various domains, for instance ranging from geophysics to biology and medicine. The provided datasets are built with the help of Gligen and the already existing NuImages, Ozfish and Deepfish datasets. The file "CarGenerated.zip" contains images generated with Gligen and with provided bounding boxes around cars in an urban environment. The file "fishes_on_bkg.zip" provides fish images generated with fishes from Deepfish inpainted with Gligen on generated backgrounds. The file "fish_text.zip" contains images completely generated with Gligen containing fishes with annotated bounding boxes. Finally, the file "oz_masked_512.zip" contains a simpler dataset of copy paste images of Deepfish fishes on Ozfish backrounds. All the files contains the images saved in different folders for training and validation, plus an index file called gt_fish.csv for the bounding boxes.

DeepFish_cls

## Overview

DeepFish_cls is a dataset for object detection tasks - it contains Objects annotations for 8,179 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [CC BY 4.0 license](https://creativecommons.org/licenses/CC BY 4.0).

Dataset used to train the segmentation, labelling and measuring estimation model based on machine learning methods in the project DeepFish2 deepfish.dtic.ua.es/

Images correspoding to the Spanish fish markets of Altea, Torrevieja and Moraira.

Information stored of the fish species, segmentation, bounding box, fish lenght (just the segment, not the real size).

This dataset gather isotopic ratios measured on 359 fish, sharks and crustaceans collected between 200 and 800 m depth, in the Mediterranean canyons during MEDITS 2012 and 2013 surveys Important Note: This submission has been initially submitted to SEA scieNtific Open data Edition (SEANOE) publication service and received the recorded DOI. The metadata elements have been further processed (refined) in EMODnet Ingestion Service in order to conform with the Data Submission Service specifications.

The dataset accompanies the DeepFish manuscript "Deep phenotypic profiling of neuroactive drugs in larval zebrafish". It consists of motion index time-series data derived from 3 phenotypic screens:

1) SCREEN-WELL Neurotransmitter Set of 650 known drugs (NT650), 2) Quality control screen of 16 known drugs (QC_Screen), and 3) ChemBridge DIVERSet Screening Library (10,000 compounds)

For each, we provide raw data files (in the form of zipped numpy arrays), and associated csv files with additional information, such as the name of the chemical and its location on the plate. The index of the row in the numpy arrays corresponds to the same index in the associated CSV files.

We also provide two saved PyTorch models from our manuscript: Twin-NN (twin-nn-saved-model-state-dict.pt.zip) and Twin-DN (twin-dn-saved-model-state-dict.pt.zip).

Finally we provide a dataset of prepared train and test pairs (as described in the manuscript) in 4 .npy arrays (pairs_train.npy, pairs_test.npy, labels_train.npy, labels_test.npy). These can be loaded by the code provided in the github repo and used to train models.

Aim: We asessed whether functional turnover in east Greenland fish driven by increased occurrence of boreal species compensates for the climate-driven loss of species thereby maintaining functional diversity and ecological function. Location: The study region encompasses a shelf and slope area located offshore between 63° N and 66° N, east of Greenland. Methods: We investigated spatio-temporal changes in fish communities based on a unique dataset covering a depth range of 1500m over 18 years combined with a demersal fish trait dataset. We analyzed the species by trait matrix using principal component analysis (PCA). To investigate trait patterns across the communities (sites), community weighted mean (CWM) traits were calculated and analysed using PCA. The CWM traits matrix was further analysed by redundancy analysis (RDA) with depth-strata and year as explanatory variables. Results: We found signs of a taxonomic and functional borealization, associated with a loss in functional diversity, down to 1000m, characterized by an increase in mobile generalists, and a decrease in bottom dwelling benthivores. The functional turnover brought about by boreal species was not sufficient to compensate for the loss of Arctic species traits, hence the loss in functional diversity. Main conclusions: The functional turnover brought about by boreal species was not sufficient to compensate for the loss of Arctic species traits that may negatively affect ecosystem robustness to environmental change. These observations are most likely not unique to this study area, and calls for the inclusion of the deep sea in climate adaptation of management strategies.

Ovaj skup podataka uključuje finansijske izvještaje, račune i blokade, te nekretnine. Podaci uključuju prihode, rashode, dobit, imovinu, obaveze i informacije o nekretninama u vlasništvu kompanije. Finansijski podaci, finansijski sažetak, sažetak kompanije, preduzetnik, zanatlija, udruženje, poslovni subjekti.