Pose Estimation with TensorFlow + openCV (pt3) 3D pose estimation

[@harryhenderson983]

Great work man! I've been looking for a tutorial similar to this one for a project I'm working on and this one really delivers!

[@SIDDHARTHRAJPATEL-YrBTechMecha]

Hey Mark, I was lifting keypoints from 2d to 3d pose , as it is not available in recent code, i have downloaded the earlier version of your code. Though i still don't know from where and how we decided the parameters in the "prob_model_params.mat" file in the "lifting" folder, to lift the poses to 3d. Can you explain please

[@teita157_kaustubhnarkhede9]

Please help, I want 2D keypoints from an image and use it in another function. Please tell me how to get those 2D keypoints.

[@胡韬-o7w]

Hi~Mark.when the mesh is run for the first time if the variable humans is empty , your code will have problems.When raising 2d pose to 2d, an error occurs if the variable humans is empty.

[@MarkJay]

韬 胡 yes. It needs a body in the first frame

[@visionindia5440]

i am getting error "estimator" module not found some one plz help me

[@intissarkhalifa5999]

hi mark, i need your help please how can i relate the joint points with lines please to see the skeleton in real time in 3d

[@MarkJay]

we go over this in the video. was it confusing?

[@sawettananjanjor4043]

Hi sir, I am having the error prob_model.py", line 140, in transform_joints assert (pose_2d.ndim == 3) AssertionError

[@henglee2859]

you can add if pose_2d_mpiis.ndim != 3: print("Nobody there"); # or return 0 if you use a function. else: //do real work now

[@hangzhou5636]

Excuse me, I can't find 'lifting folder', is that you add it later??

[@hangzhou5636]

now I know the reasion why I make mistakes, the author updated the version, someone alse can go there to download old version. thanks..

[@kathanvyas4594]

Hello sir, I am having the error "raise Exception('Not enough 2D joints identified to generate 3D pose') Exception: Not enough 2D joints identified to generate 3D pose" I saw you have written that happens because you need tohave a body in teh first frame. I have set it up so that there is a body in the first frame but it still does not detect it. The camera pops open but it does not start, there is a white screen and then it gives me this error. My all other files like run_webcam and run are working perfectly. Can you help me out in this?

[@MarkJay]

kathan Vyas hmmm, in that case, I'm not sure. Are you able to get part 1 working?

[@kathanvyas4594]

Mark Jay yes. By part one do you mean getting the 3d plane and grid? If that then yes i got the grid and all set up. Its just this part that isnt working. We have the same code for camera in run_webcam.py and even that is running but its just this file thats not working and giving me this error.

[@MarkJay]

kathan Vyas I mean the previous video in the series. It's labeled as part 1

[@kathanvyas4594]

Mark Jay yup that worked perfectly. That what makes it weird as I saw the code snippet and its mostly similar where the camera part comes. Part 1 works perfect but part 2 does not.

[@袁盛祥-r1y]

HEY Mark Jay ! I try to run pt3-webcam3D.py but i got some error : image is not defined . Could you told me how to fix the error

[@MarkJay]

蹦 蹦 sounds like you're calling an object before its defined

[@袁盛祥-r1y]

okay thanks. now i solve it but humans = self.e.inference(image, scales=[None]) have error : TypeError: inference() got an unexpected keyword argument 'scales'

[@MarkJay]

hmmm, I'm not sure why that is. I havent used this repo in a while. they may have made changes

[@mohamedamine-ne1zc]

hi sir thank you for the video ,i got error message how can i solved raise Exception('Not enough 2D joints identified to generate 3D pose') Exception: Not enough 2D joints identified to generate 3D pose please i need your help and thank you

[@MarkJay]

that happens when there isnt a body in the image. Make sure you have a body in the first camera frame

[@henglee2859]

@@MarkJay, Anyway we can exception, like if .. else.?

[@yalinie5725]

Could u tell me the name of the editor?

[@MarkJay]

Juli Nie I'm using atom. I've done s setup video on atom as well

[@intissarkhalifa5999]

please Mark where is the new version of file run.py when there are the insertion of pose 3d

[@MarkJay]

Intissar Khalifa sorry, I'm not sure what you mean

[@intissarkhalifa5999]

Mark Jay the code that you write for 3D pose where is it i think you change the file run. py of your first work??

[@MarkJay]

Intissar Khalifa that is part of the pose estimation repo. Did you get part 1 of the series working?

[@intissarkhalifa5999]

Mark Jay yes the first part it works, just if you Can help me i want to apply it in videos in my dataset that i Can extract vector of points for classification of body gesture

[@MarkJay]

Intissar Khalifa I didn't modify the run.py file. Maybe I'm still confused what info you want from the images.

[@yalinie5725]

Dear sir, new code changed something inside, such as parser.add_argument('--resolution', type=str, default='432x368', help='network input resolution. default=432x368') Then the code will be problem,how to set ' model = 'mobilenet_thin_432x368'' It will make trouble. Thanks a lot

[@MarkJay]

yes, it's good practice to specify the model.

[@yalinie5725]

Could you modify code again to make your model run with newest codes? Good luck to you..

[@themonkeyminds7252]

sir i m also getting error!!!how to sove this type of error ? ERROR*************************************************** Traceback (most recent call last): File "src/pt3-webcam3D.py", line 119, in File "src/pt3-webcam3D.py", line 49, in __init__ File "C:\Users\INFINITY\Desktop\tf-pose-estimation\src etworks.py", line 84, in model_wh width, height = map(int, resolution_str.split('x')) ValueError: invalid literal for int() with base 10: 'mobilenet_thin_432'

[@themonkeyminds7252]

WAITITNG FOR YOUR REPLY SIR:)

[@themonkeyminds7252]

SIR HOW CAN I RUN THIS SCRIPT ? PLEASE SGGEST WERE TO WIRTE CODE AND WERRE

[@mohamedamineharabi9189]

hi sir nice work , I have error message pleas can you help my : Traceback (most recent call last): File "webcam3D.py", line 118, in t = Terrain() File "webcam3D.py", line 48, in __init__ w, h = model_wh(model) File "/home/amine/tf-openpose/src/networks.py", line 84, in model_wh width, height = map(int, resolution_str.split('x')) ValueError: invalid literal for int() with base 10: 'mobilenet_thin_432' thank you in advance

[@WhyGoThere]

I have the same error. Did you find a solution??

[@WhyGoThere]

I got it to run. In the main code just define (replace previous): model = 'mobilenet_thin' self.poseLifting = Prob3dPose('./src/lifting/models/prob_model_params.mat') In networks.py (replace also): def model_wh(resolution_str): width = 432 height = 368 #width, height = map(int, resolution_str.split('x')) #if width % 16 != 0 or height % 16 != 0: # raise Exception('Width and height should be multiples of 16. w=%d, h=%d' % (width, height)) return int(width), int(height) Not the nicest solution but it works.

[@mohamedamineharabi9189]

thank you for help , but i got a new error : raise Exception('Not enough 2D joints identified to generate 3D pose') Exception: Not enough 2D joints identified to generate 3D pose do you have idea to solve it thank you

[@CrystalDataMusic]

@@WhyGoThere A nicer solution is to use w, h = model_wh('432x368') in pt3-webcam3D.py (or whatever your main file is called), since networks.py splits the string into integers separated by the string x.