手部姿态估计常用公开数据集：

手部姿态估计常用公开数据集：

1.FreiHAND：

In our recent publication we presented the challenging FreiHAND dataset, a dataset for hand pose and shape estimation from single color image, which can serve both as training and benchmarking dataset for deep learning algorithms. It contains 4*32560 = 130240 training and 3960 evaluation samples. Each training sample provides:

RGB image (224x224 pixels)
Hand segmentation mask (224x224 pixels)
Intrinsic camera matrix K
Hand scale (metric length of a reference bone)
3D keypoint annotation for 21 Hand Keypoints
3D shape annotation

The training set contains 32560 unique samples post processed in 4 different ways to remove the green screen background. Each evaluation sample provides an RGB image, Hand scale and intrinsic camera matrix. The keypoint and shape annotation is withhold and scoring of algorithms is handled through our Codalab evaluation server. For additional information please visit our project page.

2.RHD (Rendered Hand Dataset)：

This dataset has been used to train convolutional networks in our paper Learning to Estimate 3D Hand Pose from Single RGB Images.

It contains 41258 training and 2728 testing samples. Each sample provides:

RGB image (320x320 pixels)
Depth map (320x320 pixels)
Segmentation masks (320x320 pixels) for the classes: background, person, three classes for each finger and one for each palm
21 Keypoints for each hand with their uv coordinates in the image frame, xyz coordinates in the world frame and a visibility indicator
Intrinsic Camera Matrix K

It was created with freely available characters from www.mixamo.com and rendered with www.blender.org. For more details on how the dataset was created please see the mentioned paper.

3.CMU Panoptic Dataset ：

openpose，它不光实现了人脸、身体关键点的检测，同时也实现了手部关键点的检测，并提供了很多数据用于训练。

4.STB (Stereo Hand Pose Tracking Benchmark)：

provides both 2D and 3D annotations of 21 keypoints for 18000 stereo pairs with a resolution of 640 × 480. The dataset shows a single person’s left hand in front of 6 different backgrounds and under varying lighting conditions.

5.OneHand10K

6.InterHand2.6M

7.COCO-WholeBody-Hand

8.Dexter ：

Dexter is a dataset providing 3129 images showing two operators performing different kinds of manipulations with a cuboid in a restricted indoor setup. The dataset provides color images, depth maps, and annotations for fingertips and cuboid corners. The color images have a spatial resolution of 640×320. Due to the incomplete hand annotation, we use this dataset only for investigating the cross-dataset generalization of our network. We refer to this test set as Dexter.（不完整，只含有5个指尖关节点）