Official page: https://github.com/CMU-Perceptual-Computing-Lab/openpose
OpenPose would not be possible without the CMU Panoptic Studio dataset.
原文链接: https://blog.csdn.net/surserrr/java/article/details/89501491
tensorflow版本: github.com/ildoonet/tf-pose-estimation
keras版本: github.com/michalfaber/keras_Realtime_Multi-Person_Pose_Estimation
pytorch版本: github.com/Hzzone/pytorch-openpose
keras版本的模型是原论文模型转码过来的，tensorflow是自己训练的模型。
在自己的图片上简单测试了一下，keras效果最好，pytorch版本的效果最差。
但是代码是pytorch的最简洁。
不同模型的分辨率参考
Download Tensorflow Graph File(pb file)
Before running demo, you should download graph files. You can deploy this graph on your mobile or other platforms.
- cmu (trained in 656x368)
- mobilenet_thin (trained in 432x368)
- mobilenet_v2_large (trained in 432x368)
- mobilenet_v2_small (trained in 432x368)
复现CAFFE版本
A 2020 Guide for Installing OpenPose
Openpose 搭建过程 （基于Ubuntu18.04） [实践复现]
怕与Tensorflow的配置相冲突。
复现TF PY版本
Ref: 人体姿态识别--Openpose+Tensorflow
TensorRT相关问题：
From TensorFlow 1.14.1, did the switch occur. When I say switch I mean:
import tensorflow.contrib.tensorrt as trt (used in ≤ TensorFlow 1.13.1 ) -->
from tensorflow.python.compiler.tensorrt import trt (TensorFlow ≥ 1.14.1)
tf.contrib doesn't exist in 2.0.
Ubuntu 18.04 的配置参考。
$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2018 NVIDIA Corporation
Built on Sat_Aug_25_21:08:01_CDT_2018
Cuda compilation tools, release 10.0, V10.0.130
$ nvidia-smi
Sat May 16 17:28:03 2020
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 435.21 Driver Version: 435.21 CUDA Version: 10.1 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce MX250 Off | 00000000:3C:00.0 Off | N/A |
| N/A 51C P0 N/A / N/A | 285MiB / 2002MiB | 3% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| 0 1573 G /usr/lib/xorg/Xorg 156MiB |
| 0 1740 G /usr/bin/gnome-shell 58MiB |
| 0 2218 G ...AAAAAAAAAAAAAAgAAAAAAAAA --shared-files 70MiB |
+-----------------------------------------------------------------------------+
演示效果如下，切忌不可打开其他视频应用抢占GPU资源。
复现TF C++版本
迁移学习
其实就是如何 retrain to improve的问题。
https://github.com/ildoonet/tf-pose-estimation/blob/master/etcs/training.md
原始数据库：http://domedb.perception.cs.cmu.edu/
/* implement */