This repo is TensorFlow implementation of Simple Baselines for Human Pose Estimation and Tracking (ECCV 2018) of MSRA for 2D multi-person pose estimation from a single RGB image.

What this repo provides:

• TensorFlow implementation of Simple Baselines for Human Pose Estimation and Tracking.
• Flexible and simple code.
• Compatibility for most of the publicly available 2D multi-person pose estimation datasets including MPII, PoseTrack 2018, and MS COCO 2017.
• Human pose estimation visualization code (modified from Detectron).

• TensorFlow
• CUDA
• cuDNN
• Anaconda
• COCO API

This code is tested under Ubuntu 16.04, CUDA 9.0, cuDNN 7.1 environment with two NVIDIA 1080Ti GPUs.

Python 3.6.5 version with Anaconda 3 is used for development.

The ${POSE_ROOT} is described as below.

${POSE_ROOT}
|-- data
|-- lib
|-- main
|-- tool
`-- output

• data contains data loading codes and soft links to images and annotations directories.
• lib contains kernel codes for 2d multi-person pose estimation system.
• main contains high-level codes for training or testing the network.
• tool contains dataset converter. I set MS COCO as reference format and provide mpii2coco and posetrack2coco converting code.
• output contains log, trained models, visualized outputs, and test result.

You need to follow directory structure of the data as below.

${POSE_ROOT}
|-- data
|-- |-- MPII
|   `-- |-- dets
|       |   |-- human_detection.json
|       |-- annotations
|       |   |-- train.json
|       |   `-- test.json
|       `-- images
|           |-- 000001163.jpg
|           |-- 000003072.jpg
|-- |-- PoseTrack
|   `-- |-- dets
|       |   |-- human_detection.json
|       |-- annotations
|       |   |-- train2018.json
|       |   |-- val2018.json
|       |   `-- test2018.json
|       |-- original_annotations
|       |   |-- train/
|       |   |-- val/
|       |   `-- test/
|       `-- images
|           |-- train/
|           |-- val/
|           `-- test/
|-- |-- COCO
|   `-- |-- dets
|       |   |-- human_detection.json
|       |-- annotations
|       |   |-- person_keypoints_train2017.json
|       |   |-- person_keypoints_val2017.json
|       |   `-- image_info_test-dev2017.json
|       `-- images
|           |-- train2017/
|           |-- val2017/
|           `-- test2017/
`-- |-- imagenet_weights
|       |-- resnet_v1_50.ckpt
|       |-- resnet_v1_101.ckpt
|       `-- resnet_v1_152.ckpt

• In the tool, run python mpii2coco.py to convert MPII annotation files to MS COCO format (MPII/annotations).
• In the tool, run python posetrack2coco.py to convert PoseTrack annotation files to MS COCO format (PoseTrack/annotations).
• In the training stage, GT human bbox is used, and human_detection.json is used in testing stage which should be prepared before testing and follow MS COCO format.
• Download imagenet pre-trained resnet models from tf-slim and place it in the data/imagenet_weights.
• Except for annotations of the MPII and PoseTrack, all other directories are original version of downloaded ones.
• If you want to add your own dataset, you have to convert it to MS COCO format.
• You can change default directory structure of data by modifying dataset.py of each dataset folder.

You need to follow the directory structure of the output folder as below.

${POSE_ROOT}
|-- output
|-- |-- log
|-- |-- model_dump
|-- |-- result
`-- |-- vis

• Creating output folder as soft link form is recommended instead of folder form because it would take large storage capacity.
• log folder contains training log file.
• model_dump folder contains saved checkpoints for each epoch.
• result folder contains final estimation files generated in the testing stage.
• vis folder contains visualized results.
• You can change default directory structure of output by modifying main/config.py.

• Run pip install -r requirement.txt to install required modules.
• Run cd ${POSE_ROOT}/lib and make to build NMS modules.
• In the main/config.py, you can change settings of the model including dataset to use, network backbone, and input size and so on.

In the main folder, run

python train.py --gpu 0-1

to train the network on the GPU 0,1.

If you want to continue experiment, run

python train.py --gpu 0-1 --continue

--gpu 0,1 can be used instead of --gpu 0-1.

Place trained model at the output/model_dump/$DATASET/ and human detection result (human_detection.json) to data/$DATASET/dets/.

In the main folder, run

python test.py --gpu 0-1 --test_epoch 140

to test the network on the GPU 0,1 with 140th epoch trained model. --gpu 0,1 can be used instead of --gpu 0-1.

Here I report the performance of the model from this repo and the original paper. Also, I provide pre-trained models and human detection results.

As this repo outputs compatible output files for MS COCO and PoseTrack, you can directly use cocoapi or poseval to evaluate result on the MS COCO or PoseTrack dataset. You have to convert the produced mat file to MPII mat format to evaluate on MPII dataset following this.

For all methods, the same human detection results are used (download link is provided at below). For comparison, I used pre-trained model from original repo to report the performance of the original repo. The table below is APs on COCO val2017 set.

• Human detection result on val2017 (55.3 AP on human class) and test-dev2017 (57.2 AP on human class) [bbox on val2017] [bbox on test-dev2017]
• Other human detection results on val2017 [Detectron_MODEL_ZOO]

The pre-trained model on COCO dataset is used for training on the PoseTrack dataset following paper. After training model on the COCO dataset, I set lr, lr_dec_epoch, end_epoch in config.py to 5e-5, [150, 155], 160, respectively. Then, run python train.py --gpu $GPUS --continue. The table below is APs on validation set.

• Human detection result on validation set [bbox]

1. Add graph.finalize when your machine takes more memory as training goes on. [issue]

2. For those who suffer from FileNotFoundError: [Errno 2] No such file or directory: 'tmp_result_0.pkl' in testing stage, please prepare human detection result properly. The pkl files are generated and deleted automatically in testing stage, so you don't have to prepare them. Most of this error comes from inproper human detection file.

This repo is largely modified from TensorFlow repo of CPN and PyTorch repo of Simple.

[1] Xiao, Bin, Haiping Wu, and Yichen Wei. "Simple Baselines for Human Pose Estimation and Tracking". ECCV 2018.