amanbhandula / alphapose

AlphaPose Implementation in Pytorch along with the pre-trained weights

News!

This pytorch version of AlphaPose runs at 20 fps on COCO validation set (4.6 people per image on average) and achieves 71 AP!

AlphaPose

Alpha Pose is an accurate multi-person pose estimator, which is the first open-source system that achieves 70+ mAP (72.3 mAP) on COCO dataset and 80+ mAP (82.1 mAP) on MPII dataset. To match poses that correspond to the same person across frames, we also provide an efficient online pose tracker called Pose Flow. It is the first open-source online pose tracker that achieves both 60+ mAP (66.5 mAP) and 50+ MOTA (58.3 MOTA) on PoseTrack Challenge dataset.

AlphaPose supports both Linux and Windows!

Installation

Windows Version please check out doc/win_install.md

1. Get the code.

git clone -b pytorch https://github.com/MVIG-SJTU/AlphaPose.git

2. Install pytorch 0.4.0 and other dependencies.

pip install -r requirements.txt

3. Download the models manually: duc_se.pth (2018/08/30) (Google Drive | Baidu pan), yolov3-spp.weights(Google Drive | Baidu pan). Place them into ./models/sppe and ./models/yolo respectively.

Quick Start

• Input dir: Run AlphaPose for all images in a folder with:

python3 demo.py --indir ${img_directory} --outdir examples/res 

• Video: Run AlphaPose for a video and save the rendered video with:

python3 video_demo.py --video ${path to video} --outdir examples/res --save_video

• Webcam: Run AlphaPose using webcam and visualize the results with:

python3 webcam_demo.py --webcam 0 --outdir examples/res --vis

• Input list: Run AlphaPose for images in a list and save the rendered images with:

python3 demo.py --list examples/list-coco-demo.txt --indir ${img_directory} --outdir examples/res --save_img

• Note: If you meet OOM(out of memory) problem, decreasing the pose estimation batch until the program can run on your computer:

python3 demo.py --indir ${img_directory} --outdir examples/res --posebatch 30

• Getting more accurate: You can enable flip testing to get more accurate results by disable fast_inference, e.g.:

python3 demo.py --indir ${img_directory} --outdir examples/res --fast_inference False

• Speeding up: Checkout the speed_up.md for more details.
• Output format: Checkout the output.md for more details.
• For more: Checkout the run.md for more options

FAQ

Check out faq.md for faq.

Contributors

Pytorch version of AlphaPose is developed and maintained by Jiefeng Li, Hao-Shu Fang and Cewu Lu.

Citation

Please cite these papers in your publications if it helps your research:

@inproceedings{fang2017rmpe,
  title={{RMPE}: Regional Multi-person Pose Estimation},
  author={Fang, Hao-Shu and Xie, Shuqin and Tai, Yu-Wing and Lu, Cewu},
  booktitle={ICCV},
  year={2017}
}

@ARTICLE{2018arXiv180200977X,
  author = {Xiu, Yuliang and Li, Jiefeng and Wang, Haoyu and Fang, Yinghong and Lu, Cewu},
  title = {{Pose Flow}: Efficient Online Pose Tracking},
  journal = {ArXiv e-prints},
  eprint = {1802.00977},
  year = {2018}
}

License

AlphaPose is freely available for free non-commercial use, and may be redistributed under these conditions. For commercial queries, contact Cewu Lu

alphapose's Issues

you use yolo3 to detect the people from img or video?

you use yolo3 to detect the people from img or video?

坐标点的顺序

每一次检测时坐标点的顺序在变？？

the pretrained model is bases on COCO?

AlphaPose

I have to run it for multiple videos and get output in separate json file. Do any body has code for that ? and also get the output video representing keypoints in the video ?

Run on CPU

How to run this code on CPU machine.please guide me about this.Thanks

where is the resout imgs??

Input dir: Run AlphaPose for all images in a folder with:

python3 demo.py --indir ${img_directory} --outdir examples/res

Run AlphaPose for all images：TypeError: only integer tensors of a single element can be converted to an index

reason：invalid parameter in cv2.ellipse2Poly() in fn.py
raw code:

line 214 in fn.py : polygon = cv2.ellipse2Poly((int(mX),int(mY)), (int(length/2), stickwidth), int(angle), 0, 360, 1)

stickwidth is a float tensor which is not suitable for param.axes in cv2.ellipse2Poly(), so l tried to turn "stickwidth" to "int (stickwidth)",
thus promising code running. but this may affect the pattern of skeleton

Can the code be run under the Ubantu system?

Can the code be run under the Ubantu system?thank u

install ntpath

I used pip or conda install others,but ntpath failed,i dnt konw how to install it

I can only reach 0.698 mAP, instead of 0.71 on COCO ValidSet

Is there any problem? The pre-trained model is loaded and slice the first 17 outputs for COCO validation.

List of human keypoint names available?

I am looking for the names of the keypoints that this method returns, specifically 17 different keypoint names like Wrist, Ankle, Knee, etc. Is this information available somewhere?

Training AlphaPose

For retraining the model with my own dataser, I just need the h5 file of my dataset and the images?
In this case, Who I can retrain it?

I tried to copy the structure of the Coco dataset with my images and my .h5 and it doesn't seem to work.

Thanks in advance

question about nClasses

Found that the 'conv_out' in check_model named 'duc_se.pth' have shape of (batchsize, 33 , 80, 64),,keypoints only have 17, would you tell me why 33 but not 17? thanks !

${img_directory} -video ${path to video}

${img_directory} in the code of demo.py,change it to my own directoryof imgs,but has wrong

Are u Chinese？

If u are Chinese ,could u give me a contact with wechat or mail ,thanks.I am Chinese.

question about img.py and opt.py in train_sppe/src/predict

when i run train.py i meet this error:
Traceback (most recent call last):
File "C:\run-taining-alphapose\AlphaPose-master\AlphaPose-master\train_sppe\src\train.py", line 14, in
from evaluation import prediction
File "C:\run-taining-alphapose\AlphaPose-master\AlphaPose-master\train_sppe\src\evaluation.py", line 12, in
from predict.opt import opt
File "C:\run-taining-alphapose\AlphaPose-master\AlphaPose-master\train_sppe\src\predict\opt.py", line 1
../opt.py
^
SyntaxError: invalid syntax
about the opt.py in predict,please tell me how to modify opt.py and also img.py in train_sppe/src/predict?

Where is the pre-training file './Models/sppe/fast_res101_320x256.pth' ? Please it's important to me.

Terminate called without an active exception error

While I'm trying to run demo.py, the following error popping up.

# python demo.py --indir ${my_img_directory} --outdir ${my_output_directory} --save_img
Loading YOLO model..
Loading pose model from ./models/sppe/duc_se.pth
0it [00:00, ?it/s]
===========================> Finish Model Running.
===========================> Rendering remaining images in the queue...
===========================> If this step takes too long, you can enable the --vis_fast flag to use fast rendering (real-time).
terminate called without an active exception
Aborted (core dumped)

Is this problem caused by reading ./models/sppe/duc_se.pth file? Does anyone has the same problem? Any suggestions are welcome.

How can I use other detection models (like YoloX) as a detector in AlphaPose?

Can you guide me how I apply my detection model or other models like yoloX as a detector in AlphaPose. In which lines of code that you feed bounding boxes to SSTN and SPPE module?

Is it possible to run on AMD using OpenCL?

Just wondering if it is possible to get this to run on an AMD GPU using OpenCL?

Kindly tell how to run it on gpu,right now running on cpu

right now it is running on cpu but i want it to run on gpu kindly tell

能否多个视频同时进行检测？

python video_demo.py --video ${path to video} --outdir examples/res --save_video
此为单个视频进行骨骼点检测，请问能实现多个视频批量进行检测吗？

cannot import name 'int_classes' from 'torch._six'

Running webcam_ Errors encountered during demo

Thanks for your help！！