Forked from https://github.com/asanakoy/deeppose_tf

Modifications

• work with Python 3 and OpenCV 3.

• data location: ~/data

  So, make a data dir under your home folder.

  cd ~
  mkdir data

• script for "small" training set (using original LSP without extended images)

  examples/train_lsp_alexnet_imagenet_small.sh

NOTE: This is not an official implementation. Original paper is DeepPose: Human Pose Estimation via Deep Neural Networks.

This is implementation of DeepPose (stg-1).
Code includes training and testing on 2 popular Pose Benchmarks: LSP Extended Dataset and MPII Human Pose Dataset.

Performance of Alexnet pretrained on Imagenet and finetuned on LSP is close to the performance reported in the original paper.

• Python 3.X
• TensorFlow r1.0+
• OpenCV 3.X
• Chainer 1.17.0+ (for background data processing only)
• tqdm 4.8.4+

Requires around 10 Gb of free RAM.

1. Install TensorFlow

   With GPU, run this.

   pip install --upgrade tensorflow-gpu

   To use CUDA, add the following lines at the end of ~/.bashrc

   export PATH=$PATH:/usr/local/cuda/bin
   export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib64
   export LD_INCLUDE_PATH=$LD_INCLUDE_PATH:/usr/local/cuda/include

   To install Tensorflow without GPU, run this.

   pip install --upgrade tensorflow
   

2. Install OpenCV

3. Install other dependencies via pip.

   pip install --upgrade chainer numpy tqdm scipy

4. In scripts/config.py set ROOT_DIR to point to the root dir of the project.

5. To start from the existing weights, download weights of alexnet pretrained on Imagenet bvlc_alexnet.tf and put them into weights/ dir.

   mkdir weights
   cd weights
   wget -O bvlc_alexnet.tf https://hcicloud.iwr.uni-heidelberg.de/index.php/s/sNklPpCiqOYOCAz/download

• LSP dataset (1000 tran / 1000 test images)
• LSP Extended dataset (10000 train images)
• MPII dataset (use original train set and split it into 17928 train / 1991 test images)
  • Annotation
  • Images

download datasets

cd datasets
./download_lsp.sh   # to get LSP dataset
./download_mpii.sh  # to get MPII dataset
cd ..

generate csv files

export PYTHONPATH=`pwd`
python datasets/lsp_dataset.py
python datasets/mpii_dataset.py

examples/ provide several scripts for training on LSP + LSP_EXT and MPII:

• examples/train_lsp_alexnet_imagenet_small.sh to run training Alexnet on LSP using weights pretrained on Imagenet.
• examples/train_lsp_alexnet_scratch.sh to run training Alexnet on LSP + LSP_EXT from scratch
• examples/train_mpii_alexnet_imagenet.py to run training Alexnet on MPII using weights pretrained on Imagenet.
• examples/train_mpii_alexnet_scratch.py to run training Alexnet on MPII from scratch.

Example: To use bash script, run the following from the ROOT folder.

bash examples/train_lsp_alexnet_imagenet_small.sh

To run the python code, setup the PYTHONPATH and then launch the py code.

export PYTHONPATH="$(pwd):$(pwd)/scripts"
python examples/train_mpii_alexnet_imagenet.py 

All these scripts call train.py.
To check which options it accepts and which default values are set, please look into cmd_options.py.

• The network is trained with Adagrad optimizer and learning rate 0.0005 as specified in the paper.
• For training we use cropped pearsons (not the full image).
• To use your own network architecure set it accordingly in scripts/regressionnet.py in create_regression_net method.

The network wiil be tested during training and you will see the following output every T iterations:

8it [00:06,  1.31it/s]                                                                         
Step 0 test/pose_loss = 0.116
Step	 0	 test/mPCP	 0.005
Step 0 test/parts_PCP:
Head	Torso	U Arm	L Arm	U Leg	L Leg	mean
0.000	0.015	0.001	0.003	0.013	0.001	0.006
Step	 0	 test/mPCKh	 0.029
Step	 0	 test/mSymmetricPCKh	 0.026
Step 0 test/parts_mSymmetricPCKh:
Head	Neck	Shoulder	Elbow	Wrist	Hip	  Knee	Ankle
0.003	0.016	0.019	    0.043	0.044	0.028	0.053	0.003

Here you can see that PCP and PCKh scores at step (iteration) 0.
test/METRIC_NAME means that the metric was calculated on test set.
val/METRIC_NAME means that the metric was calculated on validation set. Just for sanity check on LSP I took the first 1000 images from train as validation.

pose_loss is the regression loss of the joint prediction,
mPCP is mean [email protected] score over all sticks,
parts_PCP is [email protected] score for every stick.
mRelaxedPCP is a relaxed [email protected] score, where the stick has a correct position when the average error for both joints is less than the threshold (0.5).
mPCKh is mean PCKh score for all joints,
mSymmetricPCKh is mean PCKh score for all joints, where the score for symmetric left/right joints was averaged,

To test the model use tests/test_snapshot.py.

• The script will produce [email protected] and [email protected] scores applied on cropped pearsons.
• Scores wiil be computed for different crops.
• BBOX EXTENSION=1 means that the pearson was tightly cropped,
  BBOX EXTENSION=1.5 means that the bounding box of the person was enlarged in 1.5 times and then image was cropped.

Usage: python tests/test_snapshot.py DATASET_NAME SNAPSHOT_PATH,

• where DATASET_NAME is 'lsp' or 'mpii',
• SNAPSHOT_PATH is the path to the snapshot.

  For example,

python tests/test_snapshot.py lsp out/lsp_alexnet_imagenet_small/checkpoint-150000

Results for Random initialization and Alexnet initialization from our CVPR 2017 paper Deep Unsupervised Similarity Learning using Partially Ordered Sets. Check the paper for more results using our initialization and Shuffle&Learn initialization.

If you use this code please cite the repo.

GNU General Public License