Microsoft COCO Pose Detection

## Overview

Microsoft COCO Pose Detection is a dataset for computer vision tasks - it contains Objects annotations for 4,968 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.

The MS COCO (Microsoft Common Objects in Context) dataset is a large-scale object detection, segmentation, key-point detection, and captioning dataset. The dataset consists of 328K images.

Splits: The first version of MS COCO dataset was released in 2014. It contains 164K images split into training (83K), validation (41K) and test (41K) sets. In 2015 additional test set of 81K images was released, including all the previous test images and 40K new images.

Based on community feedback, in 2017 the training/validation split was changed from 83K/41K to 118K/5K. The new split uses the same images and annotations. The 2017 test set is a subset of 41K images of the 2015 test set. Additionally, the 2017 release contains a new unannotated dataset of 123K images.

Annotations: The dataset has annotations for

object detection: bounding boxes and per-instance segmentation masks with 80 object categories, captioning: natural language descriptions of the images (see MS COCO Captions), keypoints detection: containing more than 200,000 images and 250,000 person instances labeled with keypoints (17 possible keypoints, such as left eye, nose, right hip, right ankle), stuff image segmentation – per-pixel segmentation masks with 91 stuff categories, such as grass, wall, sky (see MS COCO Stuff), panoptic: full scene segmentation, with 80 thing categories (such as person, bicycle, elephant) and a subset of 91 stuff categories (grass, sky, road), dense pose: more than 39,000 images and 56,000 person instances labeled with DensePose annotations – each labeled person is annotated with an instance id and a mapping between image pixels that belong to that person body and a template 3D model. The annotations are publicly available only for training and validation images.

COCO is a large-scale object detection, segmentation, and captioning dataset.

Note: * Some images from the train and validation sets don't have annotations. * Coco 2014 and 2017 uses the same images, but different train/val/test splits * The test split don't have any annotations (only images). * Coco defines 91 classes but the data only uses 80 classes. * Panotptic annotations defines defines 200 classes but only uses 133.

To use this dataset:

import tensorflow_datasets as tfds

ds = tfds.load('coco', split='train')
for ex in ds.take(4):
 print(ex)

See the guide for more informations on tensorflow_datasets.

https://storage.googleapis.com/tfds-data/visualization/fig/coco-2014-1.1.0.png" alt="Visualization" width="500px">

Description:

👉 Download the dataset here

This dataset has been specifically curated for cow pose estimation, designed to enhance animal behavior analysis and monitoring through computer vision techniques. The dataset is annotated with 12 keypoints on the cow’s body, enabling precise tracking of body movements and posture. It is structured in the COCO format, making it compatible with popular deep learning models like YOLOv8, OpenPose, and others designed for object detection and keypoint estimation tasks.

Applications:

This dataset is ideal for agricultural tech solutions, veterinary care, and animal behavior research. It can be used in various use cases such as health monitoring, activity tracking, and early disease detection in cattle. Accurate pose estimation can also assist in optimizing livestock management by understanding animal movement patterns and detecting anomalies in their gait or behavior.

Download Dataset

Keypoint Annotations:

The dataset includes the following 12 keypoints, strategically marked to represent significant anatomical features of cows:

Nose: Essential for head orientation and overall movement tracking.

Right Eye: Helps in head pose estimation.

Left Eye: Complements the right eye for accurate head direction.

Neck (side): Marks the side of the neck, key for understanding head and body coordination.

Left Front Hoof: Tracks the front left leg movement.

Right Front Hoof: Tracks the front right leg movement.

Left Back Hoof: Important for understanding rear leg motion.

Right Back Hoof: Completes the leg movement tracking for both sides.

Backbone (side): Vital for posture and overall body orientation analysis.

Tail Root: Used for tracking tail movements and posture shifts.

Backpose Center (near tail’s midpoint): Marks the midpoint of the back, crucial for body stability and movement analysis.

Stomach (center of side pose): Helps in identifying body alignment and weight distribution.

Dataset Format:

The data is structure in the COCO format, with annotations that include image coordinates for each keypoint. This format is highly suitable for integration into popular deep learning frameworks. Additionally, the dataset includes metadata like bounding boxes, image sizes, and segmentation masks to provide detail context for each cow in an image.

Compatibility:

This dataset is optimize for use with cutting-edge pose estimation models such as YOLOv8 and other keypoint detection models like DeepLabCut and HRNet, enabling efficient training and inference for cow pose tracking. It can be seamlessly integrate into existing machine learning pipelines for both real-time and post-processed analysis.

This dataset is sourced from Kaggle.

COCO-WholeBody dataset is the first large-scale benchmark for whole-body pose estimation. It is an extension of COCO 2017 dataset with the same train/val split as COCO.

COCO-WholeBody dataset is ONLY for research and non-commercial use. The annotations of COCO-WholeBody dataset belong to SenseTime Research, and are licensed under a Creative Commons Attribution 4.0 License.

For commercial usage of our COCO-WholeBody annotations, please contact Mr. Malon (machang[at]tetras[dot]ai) and cc Sheng Jin (jinsheng13[at]foxmail[dot]com).

We do not own the copyright of the images. Use of the images must abide by the Flickr Terms of Use. The users of the images accept full responsibility for the use of the dataset, including but not limited to the use of any copies of copyrighted images that they may create from the dataset.

Throughout the history of art, the pose—as the holistic abstraction of the human body's expression—has proven to be a constant in numerous studies. However, due to the enormous amount of data that so far had to be processed by hand, its crucial role to the formulaic recapitulation of art-historical motifs since antiquity could only be highlighted selectively. This is true even for the now automated estimation of human poses, as domain-specific, sufficiently large data sets required for training computational models are either not publicly available or not indexed at a fine enough granularity. With the Poses of People in Art data set, we introduce the first openly licensed data set for estimating human poses in art and validating human pose estimators. It consists of 2,454 images from 22 art-historical depiction styles, including those that have increasingly turned away from lifelike representations of the body since the 19th century. A total of 10,749 human figures are precisely enclosed by rectangular bounding boxes, with a maximum of four per image labeled by up to 17 keypoints; among these are mainly joints such as elbows and knees. For machine learning purposes, the data set is divided into three subsets—training, validation, and testing—, that follow the established JSON-based Microsoft COCO format, respectively. Each image annotation, in addition to mandatory fields, provides metadata from the art-historical online encyclopedia WikiArt.

This file contains the annotations for the ConfLab dataset, including actions (speaking status), pose, and F-formations.

------------------

./actions/speaking_status:

./processed: the processed speaking status files, aggregated into a single data frame per segment. Skipped rows in the raw data (see https://josedvq.github.io/covfee/docs/output for details) have been imputed using the code at: https://github.com/TUDelft-SPC-Lab/conflab/tree/master/preprocessing/speaking_status

The processed annotations consist of:

./speaking: The first row contains person IDs matching the sensor IDs,

The rest of the row contains binary speaking status annotations at 60fps for the corresponding 2 min video segment (7200 frames).

./confidence: Same as above. These annotations reflect the continuous-valued rating of confidence of the annotators in their speaking annotation.

To load these files with pandas: pd.read_csv(p, index_col=False)

./raw.zip: the raw outputs from speaking status annotation for each of the eight annotated 2-min video segments. These were were output by the covfee annotation tool (https://github.com/josedvq/covfee)

Annotations were done at 60 fps.

--------------------

./pose:

./coco: the processed pose files in coco JSON format, aggregated into a single data frame per video segment. These files have been generated from the raw files using the code at: https://github.com/TUDelft-SPC-Lab/conflab-keypoints

To load in Python: f = json.load(open('/path/to/cam2_vid3_seg1_coco.json'))

The skeleton structure (limbs) is contained within each file in:

f['categories'][0]['skeleton']

and keypoint names at:

f['categories'][0]['keypoints']

./raw.zip: the raw outputs from continuous pose annotation. These were were output by the covfee annotation tool (https://github.com/josedvq/covfee)

Annotations were done at 60 fps.

---------------------

./f_formations:

seg 2: 14:00 onwards, for videos of the form x2xxx.MP4 in /video/raw/ for the relevant cameras (2,4,6,8,10).

seg 3: for videos of the form x3xxx.MP4 in /video/raw/ for the relevant cameras (2,4,6,8,10).

Note that camera 10 doesn't include meaningful subject information/body parts that are not already covered in camera 8.

First column: time stamp

Second column: "()" delineates groups, "<>" delineates subjects, cam X indicates the best camera view for which a particular group exists.

phone.csv: time stamp (pertaining to seg3), corresponding group, ID of person using the phone

COCO formatted 2D skeletal keypoints for YouTube and Clinical datasets from: Computer vision to automatically assess infant neuromotor risk (Chambers et al 2020) extracted with ViTPose-H implemented in MMPose. For original dataset and data see: https://figshare.com/s/10034c230ad9b2b2a6a4Includes:json files with bounding boxes and 2D keypoints/confidencesvideo metadata (fps, original dimensions)Data descriptions:Youtube Dataset: 94 infants, 19 excluded19 annotations removed for meeting one or more of the following exclusion criteria:8 Partially overlapping twins4 NICU and/or hospital settings1 In water1 On a rocker2 Face occluded by caregiver and/or toy3 Low contrast/very poor video qualityClinical Dataset: 19 infants (31 videos total)Additional data available in original figshare

Because of their close relationship with humans, non-human apes (chimpanzees, bonobos, gorillas, orangutans, and gibbons, including siamangs) are of great scientific interest. The goal of understanding their complex behavior would be greatly advanced by the ability to perform video-based pose tracking. Tracking, however, requires high-quality annotated datasets of ape photographs. Here we present OpenApePose, a new public dataset of 71,868 photographs, annotated with 16 body landmarks, of six ape species in naturalistic contexts. We show that a standard deep net (HRNet-W48) trained on ape photos can reliably track out-of-sample ape photos better than networks trained on monkeys (specifically, the OpenMonkeyPose dataset) and on humans (COCO) can. This trained network can track apes almost as well as the other networks can track their respective taxa, and models trained without one of the six ape species can track the held-out species better than the monkey and human models can. Ultimately, the results of our analyses highlight the importance of large specialized databases for animal tracking systems and confirm the utility of our new ape database.

Description: The "iRodent" dataset contains rodent species observations obtained using the iNaturalist API, with a focus on Suborder Myomorpha (Taxon ID: 16). The dataset features prominent rodent species like Muskrat, Brown Rat, House Mouse, Black Rat, Hispid Cotton Rat, Meadow Vole, Bank Vole, Deer Mouse, White-footed Mouse, and Striped Field Mouse. The dataset provides manually labeled keypoints for pose estimation and segmentation masks for a subset of images using a Mask R-CNN model.
Creator: Adaptive Motor Control Lab
Data Format: COCO format
Number of Images: 443
Species: Muskrat, Brown Rat, House Mouse, Black Rat, Hispid Cotton Rat, Meadow Vole, Bank Vole, Deer Mouse, White-footed Mouse, Striped Field Mouse
Image Resolution: Varied (800x600 to 5184x3456 pixels)
Annotations: Pose keypoints and generated segmentation masks by Tian Qiu and Mackenzie Mathis.
License: Apache 2.0
Keywords: animal pose estimation, behaviour analysis, keypoints, rodent
Contact: Mackenzie Mathis
Email: mackenzie.mathis@epfl.ch

Classification performance when all spine flexion angle and patient reported outcome measures (PROMs) features were inputted.

The global coco glucoside market is experiencing robust growth, driven by increasing demand across various applications, particularly in personal care and cosmetics. The market, valued at approximately $250 million in 2025, is projected to exhibit a Compound Annual Growth Rate (CAGR) of 7% from 2025 to 2033. This growth is fueled by the rising consumer preference for natural and sustainable ingredients in personal care products, coupled with the increasing awareness of the benefits of coco glucoside as a mild, non-irritating surfactant. The versatility of coco glucoside, its biodegradability, and its effectiveness in diverse applications, including detergents, food, and medicine, further contribute to its market expansion. Leading companies like Libra Speciality Chemicals, Galaxy Surfactants, and BASF are actively investing in research and development to enhance coco glucoside production and explore new applications, driving innovation within the market. The regional distribution reflects strong growth across North America and Europe, with emerging markets in Asia-Pacific showcasing significant potential for future expansion. While regulatory changes and fluctuations in raw material prices pose potential challenges, the overall market outlook for coco glucoside remains optimistic, promising considerable growth opportunities in the coming years. The market segmentation analysis reveals personal care and cosmetics to be the dominant application segment, followed by detergents. Growth in the food and pharmaceutical segments is expected to be slower but steady, driven by increased consumer demand for natural and effective ingredients in both food processing and pharmaceutical formulations. The liquid form of coco glucoside holds a larger market share than the cream form, owing to its ease of use in various applications. Geographic expansion will primarily focus on regions with strong economic growth and developing industries such as personal care and manufacturing. This includes a focus on emerging economies within Asia-Pacific and expansion into new market segments within existing regions like North America and Europe. Competitive dynamics within the market are characterized by a mix of large multinational corporations and specialized chemical manufacturers, with focus on innovation and sustainable production methods becoming increasingly crucial for success. This comprehensive report provides an in-depth analysis of the global coco glucoside market, a rapidly expanding sector valued at over $1.5 billion in 2023. We project a Compound Annual Growth Rate (CAGR) exceeding 7% through 2030, driven by increasing demand across various industries. This report delves into market dynamics, key players, emerging trends, and future growth projections, offering invaluable insights for stakeholders seeking to navigate this lucrative market.

This dataset contains different scenarios wherein a mobile robot is approaching a set of YCB objects using both its base and arm motions. There are a total of sixteen sequences with around 100-time steps per sequence. All the sequences were generated using BlenderProc photo-realistic renderer (https://github.com/DLR-RM/BlenderProc). Eight different YCB objects were used. All these objects have a unique 6D pose, while some of the objects also have a single or multiple axes of symmetry. In each sequence maximum of three objects were randomly sampled. In addition, for each sequence, there are five views of the objects from external cameras placed between 2.5-3-5 m facing towards the objects.

This dataset was used for experimental evaluation in the following ICRA 2022 paper:

Naik, L., Iversen, T. M., Kramberger, A., Wilm, J., & Krüger, N. (Accepted/In press). Multi-view object pose distribution tracking for pre-grasp planning on mobile robots. In 2022 IEEE International Conference on Robotics and Automation (ICRA) IEEE.

Technical details:

In each sequence, the first 5 frames (0-4) contain views from external cameras while frames (5-104) provides a view of the objects visible in the robot camera as it approaches the objects. All the ground truths are provided using the 'coco' annotations format.

The 3D surgical tool dataset (3dStool) has been constructed with the aim of assisting the development of computer vision techniques that address the operating room. Note, functions for visualisation, processing, and splitting the dataset can be found in the relevant github repository.

Specifically, even though laparoscopic scenes have received a lot of attention in terms of labelled images, surgical tools that are used at initial stages of an operation, such as scalpels and scissors, have not had any such datasets developed.

3dStool includes 5370 images, accompanied by manually drawn polygon labels, as well as information on the 3D pose of these tools in operation. The tools were recorded while operating on a cadaveric knee. A RealSense D415 was used for image collection, while an optical tracker was employed for the purpose of 3D pose recording. Four surgical tools have been collected for now:

1. Scalpel
2. Scissors
3. Forceps
4. Electric Burr

An annotation json file (in the format of COCO) exists for the images, containing the masks, boxes, and other relevant information. Furthermore, pose information is provided in two different manners.

Firstly, a csv in the following format:

Position and orientation are both provided in the coordinate axes of the camera used to obtain the data (Realsense D415, Intel, USA). Pose is provided in the form of quaternions, however it is possible to convert this format into other available notations.

The pose data can also be combined with the masks in the form of a final json file, in order to obtain a final COCO-format json with object poses as well. In the data provided, each of the test, train and validation subsets have their own COCO-like json files with the poses fused within, although the "orignal_jsons" only provide the image masks.

The files and directories are structured as follows. Note that this example is based on the "train" directory, but a similar structure has been created for the test and val sets:

• Train
  • manual_json - Contains the json created when manually annotating, the images, therefore no pose data included
  • pose - Contains the CSV file with the poses of the relevant images, explained in the table above
  • pose_json - Contains the fused json that includes both the annotations and the pose data for each image
  • surgical2020 - Contains the images in jpg format

Set of optimal input features yielding the highest classification performance.

GazeFollow is a large-scale dataset annotated with the location of where people in images are looking. It uses several major datasets that contain people as a source of images: 1, 548 images from SUN, 33, 790 images from MS COCO, 9, 135 images from Actions 40, 7, 791 images from PASCAL, 508 images from the ImageNet detection challenge and 198, 097 images from the Places dataset. This concatenation results in a challenging and large image collection of people performing diverse activities in many everyday scenarios.

Abstract

This paper presents SubPipe, an underwater dataset for SLAM, object detection, and image segmentation. SubPipe has been recorded using a lightweight autonomous underwater vehicle (LAUV), operated by OceanScan MST, and carrying a sensor suite including two cameras, a side-scan sonar, and an inertial navigation system, among other sensors. The AUV has been deployed in a pipeline inspection environment with a submarine pipe partially covered by sand. The AUV's pose ground truth is estimated from the navigation sensors. The side-scan sonar and RGB images include object detection and segmentation annotations, respectively. State-of-the-art segmentation, object detection, and SLAM methods are benchmarked on SubPipe to demonstrate the dataset's challenges and opportunities for leveraging computer vision algorithms.To the authors' knowledge, this is the first annotated underwater dataset providing a real pipeline inspection scenario. The dataset and experiments are publicly available online.

On Zenodo we provide three versions for SubPipe. One is the full version (SubPipe.zip, ~80GB unzipped) and two subsamples: SubPipeMini.zip, ~12GB unzipped and SubPipeMini2.zip, ~16GB unzipped. Both subsamples are only parts of the entire dataset (SubPipe.zip). SubPipeMini is a subset, containing semantic segmentation data, and it has interesting camera data of the underwater pipeline. On the other hand, SubPipeMini2 is mainly focused on underwater side-scan sonar images of the seabed including ground truth object detection bounding boxes of the pipeline.

For (re-)using/publishing SubPipe, please include the following copyright text:

SubPipe is a public dataset of a submarine outfall pipeline, property of Oceanscan-MST. This dataset was acquired with a Light Autonomous Underwater Vehicle by Oceanscan-MST, within the scope of Challenge Camp 1 of the H2020 REMARO project.

More information about OceanScan-MST can be found at this link.

Cam0 — GoPro Hero 10

Camera parameters:

Resolution: 1520×2704

fx = 1612.36

fy = 1622.56

cx = 1365.43

cy = 741.27

k1,k2, p1, p2 = [−0.247, 0.0869, −0.006, 0.001]

Side-scan Sonars

Each sonar image was created after 20 “ping” (after every 20 new lines) which corresponds to approx. ~1 image / second.

Regarding the object detection annotations, we provide both COCO and YOLO formats for each annotation. A single COCO annotation file is provided per each chunk and per each frequency (low frequency vs. high frequency), whereas the YOLO annotations are provided for each SSS image file.

Metadata about the side-scan sonar images contained in this dataset:

Images for object detection

Low Frequency (LF):

5000

LF image size: 2500 × 500

High Frequency (HF):

5030

HF Image size 5000 × 500

Total number of images: 10030

Annotations

Low Frequency:

3163

High Frequency:

3172

Total number of annotations: 6335