Hi team, Appreciations for offering production ready pose estimation!

I understand that you have released this project under Apache License 2.0 which means that this project can be used in commercial projects for free.

Also, you have shared the license - Apache License 2.0 of Model Weights and Model Code which means, that project can also be used in commercial projects for free.

Recently, I was reading license of OpenPose and found out that they don't offer commercial license for free.

According to CMU's OpenPose License, we can't use their pretrained model which follows the same license as their code, this is the issue, I'm referring to.

Since TF Implementation of OpenPose didn't use code of the CMU's OpenPose but they have used CMU's Caffe Model and converted it to Tensorflow Model, I think this project also uses the same tensorflow model. This issue points out this.

Since this project and TF Implementation of OpenPose is a derivative of CMU's OpenPose work as both are using CMU's pretrained model, so it's owned by CMU only based on the license of openpose. Could you look this issue from OpenPose repo for more detail?

There are several implementations of OpenPose using Tensorflow and PyTorch and nearly all of these offer license for free commercial usage. But every implementation is automatically CMU's property only as per their license and licensed under theirs Caffe implementation of OpenPose only.

Does that mean, this or any other implementation of Openpose or using their pretrained model in any form can't be used commercially for free!

Do you think this project along with pretrained model can be used commercially for free? What are your views in this?

Please feel free to correct me, if I am wrong somewhere.

Thanks

In the readme we read:

To run the docker image, which automatically starts the model serving API, run:

$ docker run -it -p 5000:5000 quay.io/codait/max-human-pose-estimator

This will pull a pre-built image from the Quay.io container registry (or use an existing image if already cached locally) and run it. If you'd rather checkout and build the model locally you can follow the run locally steps below.

But trying this command ends with:

docker run -it -p 5000:5000 max-human-pose-estimator
Unable to find image 'max-human-pose-estimator:latest' locally
docker: Error response from daemon: pull access denied for max-human-pose-estimator, repository does not exist or may require 'docker login': denied: requested access to the resource is denied.
See 'docker run --help'.

After registering an account and successfully using docker login the same message appears.

the Run the Notebook section of the README.md indicates:

the demo requires jupyter, numpy, matplotlib, cv2, json, and requests

however, to get cv2 i had to run pip install opencv-python. i tried just pip install cv2 but that failed. it may be worth noting/clarifying that the demo requires opencv-python (to get cv2).

currently the /predict API returns only line segments, but it would be more useful if it also included position and detail for the various body parts detected (i.e., part_id, score, x, y, part_name). for example:

{
  "status": "ok",
  "predictions": [
    {
      "human_id": 0,
      "pose_lines": [
        {
          "line": [
            308,
            417,
            479,
            359
          ]
        }
      ],
      "body_parts": [
        {
          "part_id": 5,
          "part_name": "LShoulder",
          "score": 0.1944918930530548,
          "x": 479,
          "y": 359
        }
      ]
    }
  ]
}

It just says file for the /model/predict endpoint

See #21 which uncovered a potentially fatal NameError. Add flake8 check to CI.

Hi,

I'm using this API to detect human pose on many images (5-10 FPS), but it's very difficult to keep this speed to analysis without GPU support. (I have a machine with Tesla M60 GPU and i can't use it with this API)

When do you plan to deploy GPU usage ?
I will try on my side to find edits to adapt the code with GPU usage

Thanks for your help,

Kind regards,
Florian

Human-Pose-Estimator has the same problem as Facial Age Estimator. Image Caption Generator, Facial Recognizer, and Object Detector support both JPG and PNG files. On the other hand, Human-Pose-Estimator and Facial Age Estimator can handle only a JPG file. It will be easy for developers to switch MAX container which is related to image recognition if all containers support both JPG and PNG files.

Kubernetes deployment command points to GHE:

$ kubectl apply -f https://github.ibm.com/CODAIT/MAX-Human-Pose-Estimator/raw/master/max-human-pose-estimator.yaml

Build the model cloning instruction points to GHE:

$ git clone https://github.ibm.com/ibm/MAX-Human-Pose-Estimator.git

Receiving this input

/mode/predict returns HTTP code 500 and the console output contains a stack trace entry
similar to the following: ValueError: Cannot feed value of shape (1, 615, 461, 4) for Tensor 'pnet/input:0', which has shape '(?, ?, ?, 3)'

I am complimenting this human pose estimation on my local Raspberry PI. Since the docker image and the base docker image are both made for x86 processors, I cannot use them and so I am just preparing the environment myself. After installing all requirements and other basic tools, I got the error: "No module named '_pafprocess'
you need to build c++ library for pafprocess.
See: https://github.com/ildoonet/tf-pose-estimation/tree/master/tf_pose/pafprocess"
Based on what I read it is a low-level tf library. Is there a way to solve this issue and run the program?

Hi,

I'm using the Human-pose-estimator for my tests (building image locally), and for some reasons, I need to change my machine to a Jetson Nano.

The main problem is that the DockerFile is for x86 machines and Jetson nano running on aarch64. Do you have any image for this architecture, to build this image on my Jetson Nano ?

Thanks a lot,

Regards,
Florian