Openpose from CMU implemented using Tensorflow. It also provides several variants that have made some changes to the network structure for real-time processing on the CPU.
Original Repo(Caffe) : https://github.com/CMU-Perceptual-Computing-Lab/openpose
Features
-
CMU's original network architecture and weights.
-
Transfer Original Weights to Tensorflow
-
Training Code with multi-gpus
-
-
Post processing from network output.
-
Accurate Result
-
Fast Computation
-
-
Faster network variants using mobilenet, lcnn architecture.
-
Depthwise Separable Convolution Version
-
Mobilenet Version
-
LCNN Version
-
-
Demos
-
Realtime Webcam Demo
-
Image/Video File Demo
-
-
ROS Support.
You need dependencies below.
-
python3
-
tensorflow 1.3
-
opencv3
-
protobuf
-
python3-tk
$ git clone https://www.github.com/ildoonet/tf-openpose
$ cd tf-openpose
$ git lfs pull
$ pip3 install -r requirements.txt
Dataset | Model | Description | Inference Time 1 core cpu |
---|---|---|---|
Coco | cmu | CMU's original version. Same network, same weights. | 368x368 @ 10s 320x240 @ 3.65s |
Coco | dsconv | Same architecture as the cmu version except for the depthwise separable convolution of mobilenet. |
368x368 @ 1.1s 320x240 @ 0.44s |
Coco | mobilenet | Feature extraction layers is replaced from VGG to Mobilenet from Google |
- Test being processed. This will be updated soon.
CMU Perceptual Computing Lab has modified Caffe to provide data augmentation. See : https://github.com/CMU-Perceptual-Computing-Lab/caffe_train
I implemented the augmentation codes as the way of the original version, See pose_dataset.py and pose_augment.py. This includes scaling, rotation, flip, cropping.
This process can be a bottleneck for training, so if you have enough computing resources, please see Run for Faster Training Section
$ python3 train.py --model=cmu --datapath={datapath} --batchsize=64 --lr=0.001 --modelpath={path-to-save}
2017-09-27 15:58:50,307 INFO Restore pretrained weights...
If you have enough computing resources in multiple nodes, you can launch multiple workers on nodes to help data preparation.
worker-node1$ python3 pose_dataworker.py --master=tcp://host:port
worker-node2$ python3 pose_dataworker.py --master=tcp://host:port
worker-node3$ python3 pose_dataworker.py --master=tcp://host:port
...
After above preparation, you can launch training script with special arguments.
$ python3 train.py --remote-data=tcp://0.0.0.0:port
2017-09-27 15:58:50,307 INFO Restore pretrained weights...
Also, You can quickly train with multiple gpus. This automatically splits batch into multiple gpus for forward/backward computations.
$ python3 train.py --remote-data=tcp://0.0.0.0:port --gpus=8
2017-09-27 15:58:50,307 INFO Restore pretrained weights...
I trained models within a day with 8 gpus and multiple preprocessing nodes with 48 core cpus.
[1] https://github.com/CMU-Perceptual-Computing-Lab/openpose
[2] Training Codes : https://github.com/ZheC/Realtime_Multi-Person_Pose_Estimation
[3] Custom Caffe by Openpose : https://github.com/CMU-Perceptual-Computing-Lab/caffe_train
[1] Original Paper : https://arxiv.org/abs/1704.04861
[2] Pretrained model : https://github.com/tensorflow/models/blob/master/slim/nets/mobilenet_v1.md
[1] Tensorpack : https://github.com/ppwwyyxx/tensorpack