geaxgx / depthai_blazepose Goto Github PK

Hi, i'm trying to run and test your repository, but every time I launch this python3 demo.py -i path/to/video.mp4 , I get the following error:

Pose detection blob file : depthai_blazepose/models/pose_detection_sh4.blob
Landmarks using blob file : depthai_blazepose/models/pose_landmark_full_sh4.blob
Video FPS: 30
Original frame size: 852x480
Padding on height : 186
Frame working size: 852x480
2254 anchors have been created
Creating pipeline...
Creating Pose Detection Neural Network...
Creating Landmark Neural Network...
Pipeline created.
Traceback (most recent call last):
File "demo.py", line 46, in
pose = BlazeposeDepthai(input_src=args.input,
File "depthai_blazepose/BlazeposeDepthai.py", line 175, in init
self.device = dai.Device(self.create_pipeline())
RuntimeError: No available devices

Can you explain me why this error appear?
Thank you,
Bests

I found a program using lazepose:https://github.com/Atzingen/QuickiumGymI
I download the requirements as the requirements.yml follows,when I try to run the quickium_solution.py,it happened like this :
Pose detection blob file : /home/ycj/QuickiumGym-main/scores/blazepose/models/pose_detection_sh4.blob
Landmarks using blob file : /home/ycj/QuickiumGym-main/scores/blazepose/models/pose_landmark_full_sh4.blob
Internal camera image size: 576 x 324 - pad_h: 126
Creating pipeline...
Creating Color Camera...
Traceback (most recent call last):
File "quickium_solution.py", line 105, in
pose = BlazeposeDepthai()
File "/home/ycj/QuickiumGym-main/scores/blazepose/BlazeposeDepthaiEdge.py", line 98, in init
self.device = dai.Device(self.create_pipeline())
File "/home/ycj/QuickiumGym-main/scores/blazepose/BlazeposeDepthaiEdge.py", line 143, in create_pipeline
manager_script.setScriptData(self.build_manager_script())
AttributeError: 'depthai.node.Script' object has no attribute 'setScriptData'

what could be the problem? should I replace the whole "blazepose" folder in that program with yours ?thank you very much!

Hi, I'm trying to include your library in my Bachelor project, but as I'm trying to understand it correctly a question appear to me.

When the xyz mode is enable, it create de stereo depth capture by linking right and left capture, and as I can see it give us directly the correct measure for the center of the hips or the shoulder. But, when I'm printing landmarks, I can see that there is also 3 output value by landmark, even if the xyz mode is disable. It is tangible data for landmark reprensentation in 3d in pixel, even if the xyz is disable ?

Thanks in advance for your answer

Hi,
I would like to do very low latency real time skeleton tracking.
is the FPS limited?
Does anyone know if it's possible to have like 30 or 60 fps? Maybe with a powerfull CPU or GPU?

Hello, I receive the error "depthai.Pipeline object has no attribute "create". Any ideas where the problem is? Thank you for your answer!

Pose detection blob file : C:\Users\Admin\depthai_blazepose-main\models\pose_detection_sh4.blob
Landmarks using blob file : C:\Users\Admin\depthai_blazepose-main\models\pose_landmark_full_sh4.blob
Internal camera FPS set to: 8
Sensor resolution: (1920, 1080)
Internal camera image size: 1792 x 1008 - crop_w:0 pad_h: 392
2254 anchors have been created
Creating pipeline...
Creating Color Camera...
Creating Pose Detection pre processing image manip...
Traceback (most recent call last):
File "C:\Users\Admin\depthai_blazepose-main\examples\semaphore_alphabet\demo.py", line 51, in
pose = BlazeposeDepthai(input_src=args.input, lm_model=args.model)
File "../..\BlazeposeDepthai.py", line 218, in init
self.device.startPipeline(self.create_pipeline())
File "../..\BlazeposeDepthai.py", line 287, in create_pipeline
pre_pd_manip = pipeline.create(dai.node.ImageManip)
AttributeError: 'depthai.Pipeline' object has no attribute 'create'
[Finished in 3.5s with exit code 1]
[shell_cmd: python -u "C:\Users\Admin\depthai_blazepose-main\examples\semaphore_alphabet\demo.py"]
[dir: C:\Users\Admin\depthai_blazepose-main\examples\semaphore_alphabet]
[path: C:\Program Files (x86)\Common Files\Oracle\Java\javapath;C:\ProgramData\DockerDesktop\version-bin;C:\Program Files\Docker\Docker\Resources\bin;C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\bin;C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\libnvvp;C:\Program Files (x86)\Common Files\Intel\Shared Libraries\redist\intel64\compiler;C:\Windows\system32;C:\Windows;C:\Windows\System32\Wbem;C:\Windows\System32\WindowsPowerShell\v1.0;C:\Windows\System32\OpenSSH;C:\Program Files (x86)\NVIDIA Corporation\PhysX\Common;C:\Program Files (x86)\Brackets\command;C:\Program Files\NVIDIA Corporation\NVIDIA NvDLISR;C:\Program Files\Git\cmd;C:\Program Files (x86)\Intel\Intel(R) Management Engine Components\DAL;C:\Program Files\Intel\Intel(R) Management Engine Components\DAL;C:\WINDOWS\system32;C:\WINDOWS;C:\WINDOWS\System32\Wbem;C:\WINDOWS\System32\WindowsPowerShell\v1.0;C:\WINDOWS\System32\OpenSSH;C:\Program Files\dotnet;C:\Program Files\OpenNI\Bin64;C:\Program Files\Nuitrack\nuitrack\nuitrack\bin;C:\Users\Admin\AppData\Local\Programs\Python\Python37\Scripts;C:\Users\Admin\AppData\Local\Programs\Python\Python37;C:\Users\Admin\AppData\Local\Microsoft\WindowsApps;"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\bin;C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\libnvvp;C:\ProgramData\DockerDesktop\version-bin;C:\Program Files\Docker\Docker\Resources\bin;C:\Program Files (x86)\Common Files\Intel\Shared";C:\ProgramData\NVIDIA Corporation\Downloader\PostProcessing\GFE\a1b8f95e233fed392281175573cfcd07\GFExperience.NvStreamSrv\amd64\server;C:\Program Files\NVIDIA Corporation\NvStreamSrv;;C:\Users\Admin\AppData\Local\Programs\Microsoft VS Code\bin;C:\Users\Admin.dotnet\tools]

i cant run the edge version on my oak d lite, because this error occurs

Hello.

I know you might not have a POE device. But i was wondering if you know how i can make you code run without the need of a terminal window being open.

Right now i cd to my project repo and call python3.8 demo_osc.py -e --oscIP 10.100.0.101 --oscPort 12345 -xyz
I have modified your code already to not show a live video, not any window other than the terminal window.
I added code to send the skeleton data over UPD / OSC.

Thanks for your advice.

self.device = dai.Device(self.create_pipeline())
RuntimeError: No available devices

On a Raspberry Pi4/b Ubuntu 22.04: Running python3 demo.py -e -xyz -3 world generates: File "/home/ubuntu/depthai_blazepose/o3d_utils.py", line 119, in init opt.background_color = np.asarray(bg_color) "AttributeError: 'None Type' object has no attribute 'background_color.

BTW: Running python3 demo.py -e -xyz without error required downgrading depthai to version 2.20.2 from 2.21.2.

Hi, again :x,

I would like to know, if there's an easy way to modify your script in order to replace the z value generated by the model, with the z value perceived by stereo capture. At the moment, I've implemented a queue with the stereo output in, but the number of request on the pipeline for the depth slowed the retrieval of the body data.

Hi sorry to bother again.
Thank you for your previous help and quick replies, now --show_3d is working amazingly on my Mac and I am very happy about it.
Now I tried to run it on my rasberry pi 4B. I have opencv and all the depthai dependencies installed. I also installed your requirements.txt file.

However when running the BlazeposeDepthai.py script with -3 / --show_3d I get the following error:

python3 BlazeposeDepthai.py -3 /home/pi/.virtualenvs/cv-installatie/lib/python3.7/site-packages/open3d/__init__.py:21: UserWarning: failed to import module _transformations warnings.warn('failed to import module %s' % name) 896 anchors have been created Traceback (most recent call last): File "BlazeposeDepthai.py", line 610, in <module> internal_fps=args.internal_fps) File "BlazeposeDepthai.py", line 166, in __init__ self.vis3d = o3d.visualization.Visualizer() AttributeError: module 'open3d' has no attribute 'visualization'

If I run it without it works fine. Is this not possible on a rasberry pi? or am I missing something?
Thank you for your time and this great Repo!

Your current example uses cam = pipeline.create(dai.node.ColorCamera)
I wonder if you can help me take advantage of the IR laser projector to improve the depth readings.

I know Blazepose used 2D images to infer the depth. I am hoping to use the OAK-D Pro POE at light. This means I should at least use the IR LED to illuminate the scene. And even better use the IR laser dot projector to sample the depths.

Thanks,
Stephan.

My camera has this problem after running for about 5 hours, and the program is directly stuck, and the program has not been changed. I suspect that the reason is that the temperature of my camera is too high, but this error is that when there is no one for a long time, someone suddenly starts to jam. Have you ever encountered this situation? I hope you can reply to me as soon as possible. Thank you.

Hello.
Thanks for making this great project.

I got it all running and am now trying to iterate through all landmarks in order to send them out the network over OSC.
I am able to send xyz

 def send_pose(self, client: udp_client, body):
              
        if body is None:
            client.send_message(OSC_ADDRESS, 0)
            return

        # create message and send
        builder = OscMessageBuilder(address=OSC_ADDRESS)
        builder.add_arg(1)
        for i, oneLM in enumerate(body.landmarks):
            if self.is_present(body, i):
#                print ("lm",i, "x", oneLM[0], "y", oneLM[1], "z", oneLM[2])
                builder.add_arg(i, arg_type='i'
                builder.add_arg(oneLM[0], arg_type='i')
                builder.add_arg(oneLM[1], arg_type='i')
                builder.add_arg(oneLM[2], arg_type='i')
                # visibility
                # builder.add_arg(x_y_z_v[3])
                
        msg = builder.build()
        client.send(msg)

But I was oping to also include the visibility score for each landmark.

In your code I see

        body.visibility = 1 / (1 + np.exp(-lm_raw[:,3]))
        body.presence = 1 / (1 + np.exp(-lm_raw[:,4]))

But I think this only refers to the full body not each landmark. ?

if I do

for i, oneBody in enumerate(body.landmarks):

I get this error TypeError: 'Body' object is not iterable

thanks for any advice.

I want to replace the Colorcamera with XLinkIn node.

I know that it will not be on edge, but I'm trying to use this pipeline since it is optimized and works on high FPS rate.

I've tried to send data through the XLinkIn node in the following way:

frame_time = time.monotonic()
frame_nn = dai.ImgFrame()
frame_nn.setTimestamp(frame_time)
frame_nn.setData(to_planar(video_frame, (video_frame.shape[1], video_frame.shape[0])))
frame_nn.setData(video_frame)
self.q_image_in.send(frame_nn)       

But it gives me this error:
[14442C10310D57D700] [1.4] [180.825] [system] [critical] Fatal error. Please report to developers. Log: 'Fatal error on MSS CPU: trap: 2A, address: 8009C784' '0'

Hi,

I've been exploring your custom BlazePose implementation and have a question on L398 and L399 in

A couple of lines down on L408 and L409 (in the case of rotation != 0) the following assignments are given:

x_shift = (w * width * shift_x * cos(rotation) - h * height * shift_y * sin(rotation)) 
y_shift = (w * width * shift_x * sin(rotation) + h * height * shift_y * cos(rotation)) 

These 0 terms result in x_shift and y_shift to be zero as well.

Here are (I believe) the defaults set by mediapipe, https://github.com/google/mediapipe/blob/ecb5b5f44ab23ea620ef97a479407c699e424aa7/mediapipe/calculators/util/rect_transformation_calculator.proto - 5 and 6 seems a bit high. Does this mean that no shift is necessary in the BlazePose implementation of DepthAI?

All the requirements seemed to have installed OK, and running the basic python3 BlazeposeDepthai.py appears to work, although my OAK-D is at present not mounted where I can get far enough away for a full length body picture (I'll need a USB3 extension cable), I do see some segments being tracked.

But running the python3 BlazeposeDepthai.py -3 I get this error:
~/depthai_blazepose-main$ python3 BlazeposeDepthai.py -3 896 anchors have been created [Open3D WARNING] GLFW Error: GLX: Failed to create context: BadValue (integer parameter out of range for operation) [Open3D WARNING] Failed to create window Traceback (most recent call last): File "BlazeposeDepthai.py", line 591, in <module> ht = BlazeposeDepthai(input_src=args.input, File "BlazeposeDepthai.py", line 165, in __init__ opt.background_color = np.asarray([0, 0, 0]) AttributeError: 'NoneType' object has no attribute 'background_color'
Suggestions?
I've zero experience with the open-3d python package. I'm running on Ubuntu 20.04 with all updates as of a few hours ago.

(I asked on the depthai forums as well, but was also curious about your take!)

I am trying to use Oak-D to get spatial data of the pose landmarks. For example, XYZ data of all the points that mark the joints.

I think the way Oak-D works for spatial detection on objects is that it draws a box around an object, and then averages out all of the xyz data of the points in the box from a depth map (created by using disparity matching), is this correct?

If so, what happens when a box is drawn around a person, and then some of the points in the box are objects that are further behind the person (such as the wall)?

In the case of getting spatial data for the body landmarks, do you think it's wiser to draw boxes around the landmarks and then average out the depth data? Or just get the depth data precisely at the landmark points?

Thanks in advance,
Jae

Hi,
Is it possible to retrain the model with custom dataset.

I am able to display the right mono camera and can see the IR LED is illuminating the scene.
But the blazepose seems to be run on the RGB camera.
I would like to use the camera in a dark environment and would like to learn how to let blazepose use the mono camera that can see in the dark thanks to the IR LED.

Hi!

I'm from ReDrawing team for the OpenCV AI 2021 competition, and we used a part of this repository code. We noticed that the repository don't have any license, so we can't derive from it, technically not even run it.

If not on purpose, could you please add a license? We also used your
depthai_hand_tracker project. Here some help about licenses.

Our project repository: link

Hi,
first of all thank you for your great work!

I got a problem that I'm not able to explain. I'm working with an OAK-D-PRO FF.

If I use the edge mode specifying also the xyz flag everything is fine, but if I just launch python3 demo.py -e (without -xyz) I always get this error:

[18443010E1A6AB0F00] [3.6] [7.446] [system] [critical] Fatal error. Please report to developers. Log: 'ResourceLocker' '358'
Traceback (most recent call last):
File "/home/user/Development/body/depthai_blazepose/depthai_blazepose/demo.py", line 65, in
frame, body = tracker.next_frame()
File "/home/user/Development/body/depthai_blazepose/depthai_blazepose/BlazeposeDepthaiEdge.py", line 486, in next_frame
res = marshal.loads(self.q_manager_out.get().getData())
RuntimeError: Communication exception - possible device error/misconfiguration. Original message 'Couldn't read data from stream: 'manager_out' (X_LINK_ERROR)'

Do you have any idea?

Thank you!

Hello,

So I am trying to run this project on my Oak-D camera, my main goal is to extract potentially data for all joints/landmarks in 3D. However I am experiences some a drop in FPS rate and glichiness whenever a skeleton appears. I don't think it is a processing power issue.

I'm working on a logic fix to make it compatible with the latest depthai package (2.7.1.0). I'd like to issue a pull request to your repository if I can. 😃
PINTO0309@d5cd3f9

I've already got everything working properly except Edge-only, but I'm wondering how to fix the following part. If you know of a way to do this, could you please let me know?

The reason why I want to fix this is because another engineer reported an issue that the latest OAK-D does not work properly with depthai==2.3.0. I feel that I want everyone in the world to enjoy your wonderful repository.

• Discord issue
  https://discord.com/channels/790680891252932659/799407361986658354/865952356659494932
  https://discord.com/channels/790680891252932659/799407361986658354/866090959175024640
  https://discord.com/channels/790680891252932659/799407361986658354/866092002923511808

I have successfully gotten it to work with 2.7.1.0 if it is not FullEdge.

Hi, I have this problem when I run your demo. The following picture shows that when I run the demo, the landmarks have a great shift from where they ought to be.

However, if I stream the image to the host and use mediapipe solution directly, the result is good.

I wonder if this problem is related to the blob file generated from tflite? How can I solve this problem?
Thanks.

I'm trying to implement a version of the DepthAI Blazepose project integrated with Yolo in order to pass multiple crops.
The problem is that if I walk away from the camera for 2 meters then the body score is 0 or very low.
I've noticed that the frame for the landmark regression is very small.

And a lot of times the cropped frame with warping is hidden in the corner like this:

I'm referring to that piece of code:

frame_nn = mpu.warp_rect_img(body.rect_points, square_frame, lm_input_length, lm_input_length)
# print(f"Body rect points: {body.rect_points} - Dim cropped frame: {cropped_frame.shape}")

cv2.imshow("Input landmark", frame_nn)

# cv2.imwrite("debug_pre_lm_yolo.png", frame_nn)

frame_nn = frame_nn / 255.

# cv2.imshow("landmark input", frame_nn)
nn_data = dai.NNData()
nn_data.setSequenceNum(pose_detection.getSequenceNum())
nn_data.setTimestamp(body_time)
nn_data.setLayer("input_1", to_planar(frame_nn, (lm_input_length, lm_input_length)))
q_landmark_in.send(nn_data)

How can be solved this problem?

Hi, I'm using this repository to test the pose estimation on an OAK-IOT device for the edge-programming and not host-programming. It works perfectly for one person but it can't detect other people. Is there any setting or way to do this?

I was trying to run BlazeposeDepthai.py either with webcam or filename on Window but each time I get error.
Can you help?

Creating pipeline...
Creating Pose Detection Neural Network...
Creating Landmark Neural Network...
Pipeline created.
Traceback (most recent call last):
  File "D:/Users/Downloads/depthai_blazepose-main/depthai_blazepose-main/BlazeposeDepthai.py", line 604, in <module>
    ht.run()
  File "D:/Users/Downloads/depthai_blazepose-main/depthai_blazepose-main/BlazeposeDepthai.py", line 408, in run
    device = dai.Device(self.create_pipeline())
RuntimeError: No available devices```

Hi, I come back to you for a simple question. I'm trying to add to your library the possibility to have the approximate position in millimeters of a specific point as proposed here https://docs.luxonis.com/projects/api/en/latest/samples/SpatialDetection/spatial_location_calculator/

But when I try it I've got non realistic value for x and y. As I understand the value are calculated compared to the center of the frame. But I've got too similarly values and I don't know how to trouble shot it. Have you done something special for making it work for the center hips or shoulders ?

I'm trying with this two point :

For each point their real position in frame are : [1149, 842] and [1256, 405]

And my spatial locator give me the respective true positions :
[-252.84371948242188, -120.76118469238281, 1234.7139892578125] and§ [-255.68605041503906, -121.71033477783203, 1234.7718505859375]

For the z position it seems to be legit, cause far away of the camera I go bigger the distance is..

I don't know if you will be able to help me

Getting the following error on my M1 Mac with the OAK-D S2 camera:

(venv) ~/Luxonis/depthai_blazepose ⎇ (main)
✗ ➜  python3 demo.py -e
Pose detection blob file : /Users/liuyue/Luxonis/depthai_blazepose/models/pose_detection_sh4.blob
Landmarks using blob file : /Users/liuyue/Luxonis/depthai_blazepose/models/pose_landmark_full_sh4.blob
Internal camera FPS set to: 20
Internal camera image size: 1152 x 648 - pad_h: 252
Creating pipeline...
Creating Color Camera...
Creating Pose Detection pre processing image manip...
Creating Pose Detection Neural Network...
Creating Pose Detection post processing Neural Network...
Creating Landmark pre processing image manip...
Creating DiveideBy255 Neural Network...
Creating Landmark Neural Network...
Pipeline created.
[19443010C1351C1300] [1.1.4] [1.199] [NeuralNetwork(5)] [warning] Network compiled for 1 shaves, maximum available 13, compiling for 6 shaves likely will yield in better performance
[19443010C1351C1300] [1.1.4] [1.200] [NeuralNetwork(9)] [warning] Network compiled for 4 shaves, maximum available 13, compiling for 6 shaves likely will yield in better performance
[19443010C1351C1300] [1.1.4] [1.433] [NeuralNetwork(4)] [warning] Network compiled for 4 shaves, maximum available 13, compiling for 6 shaves likely will yield in better performance
[19443010C1351C1300] [1.1.4] [1.446] [NeuralNetwork(5)] [warning] The issued warnings are orientative, based on optimal settings for a single network, if multiple networks are running in parallel the optimal settings may vary
Pipeline started - USB speed: SUPER
[19443010C1351C1300] [1.1.4] [1.446] [NeuralNetwork(9)] [warning] The issued warnings are orientative, based on optimal settings for a single network, if multiple networks are running in parallel the optimal settings may vary
[19443010C1351C1300] [1.1.4] [1.446] [NeuralNetwork(4)] [warning] The issued warnings are orientative, based on optimal settings for a single network, if multiple networks are running in parallel the optimal settings may vary
[19443010C1351C1300] [1.1.4] [7.415] [system] [critical] Fatal error. Please report to developers. Log: 'ResourceLocker' '358'
libc++abi: terminating due to uncaught exception of type std::runtime_error: Communication exception - possible device error/misconfiguration. Original message 'Couldn't read data from stream: 'manager_out' (X_LINK_ERROR)'
Stack trace (most recent call last):
#10   Object "libc++.1.dylib", at 0x180d313ef, in std::rethrow_exception(std::exception_ptr) + 23
#9    Object "libc++abi.dylib", at 0x180dbfeeb, in std::terminate() + 55
#8    Object "libc++abi.dylib", at 0x180dbff47, in std::__terminate(void (*)()) + 15
#7    Object "libobjc.A.dylib", at 0x180a8703b, in _objc_terminate() + 159
#6    Object "libc++abi.dylib", at 0x180db03b3, in demangling_terminate_handler() + 319
#5    Object "libc++abi.dylib", at 0x180dc0b83, in abort_message + 131
#4    Object "libsystem_c.dylib", at 0x180d15ae7, in abort + 179
#3    Object "libsystem_pthread.dylib", at 0x180e07c27, in pthread_kill + 287
#2    Object "libsystem_platform.dylib", at 0x180e36a83, in _sigtramp + 55
#1    Object "depthai.cpython-310-darwin.so", at 0x289cff467, in backward::SignalHandling::sig_handler(int, __siginfo*, void*) + 19
#0    Object "depthai.cpython-310-darwin.so", at 0x289cff4bf, in backward::SignalHandling::sig_handler(int, __siginfo*, void*) + 107
[1]    75155 abort      python3 demo.py -e

Does this repository support Apple Silicon macs?

Hi, I needed to update requirements.txt to get blazepose running -

markupsafe == 2.0.1
werkzeug == 2.0.3

Where is the place in the code that do this processing?

Hi, how would I go about determining the absolute 3d location of a key point besides the refrence point? Right now, I'm able to find what the 3d positions of the mid point between my hips or shoulders are, but I haven't been able to figure out a way to do this for say, my right hand, or any other arbitrary body part. For my project, I'm able to assume certain keypoints such as the hand are in frame, so I'm not sure if the points mentioned in the readme would still be an issue.

Script returns error:
self.device = dai.Device()
RuntimeError: Failed to find device after booting, error message: X_LINK_DEVICE_NOT_FOUND

However, the same code works on OAK-D W with IMX378 sensor.

I wonder if it's related to a setup script.

Hi thank you so much for sharing this amazing demo!

maybe its something I am doing wrong, but I cant seem to get --show_3d to work.
When I run it, it gives this error:
Traceback (most recent call last): File "/Users/corlangerak/depthai_blazepose/BlazeposeDepthai.py", line 626, in <module> ht.run() File "/Users/corlangerak/depthai_blazepose/BlazeposeDepthai.py", line 412, in run device = dai.Device(self.create_pipeline()) File "/Users/corlangerak/depthai_blazepose/BlazeposeDepthai.py", line 213, in create_pipeline pd_nn.setBlobPath(str(Path(self.pd_path).resolve().absolute())) RuntimeError: NeuralNetwork node | Blob at path: /Library/Frameworks/Python.framework/Versions/3.8/Resources/Python.app/Contents/Resources/models/pose_detection.blob doesn't exist
Am I missing something? the path is actually defined in the code, so I do not really understand.
Thanks again!

Hello @geaxgx and thank you so much for this terrific software! I have been able to get depthai_blazepose to work in many modes with my OAK-D camera. For example when I run python demo.py --edge --lm_m='lite' I see 26 FPS which is awesome!

However, I really want to be able to see the xyz coordinates, however when I run python demo.py --xyz I get several errors (see below) including a RuntimeWarning and a TypeError, and the program quits. Am I doing something wrong?

Pose detection blob file : /Volumes/GoogleDrive-102536951206467607597/My Drive/Projects/Punchout/SkeletalTracker/depthai_blazepose/models/pose_detection_sh4.blob
Landmarks using blob file : /Volumes/GoogleDrive-102536951206467607597/My Drive/Projects/Punchout/SkeletalTracker/depthai_blazepose/models/pose_landmark_full_sh4.blob
Internal camera FPS set to: 13
Sensor resolution: (1920, 1080)
Internal camera image size: 1152 x 648 - crop_w:0 pad_h: 252
2254 anchors have been created
Creating pipeline...
Creating Color Camera...
Creating Pose Detection pre processing image manip...
Creating Pose Detection Neural Network...
Creating Landmark Neural Network...
Pipeline created.
[14442C1001FA60D700] [61.040] [NeuralNetwork(12)] [warning] Number of inference threads assigned for network is 1, assigning 2 will likely yield in better performance
[14442C1001FA60D700] [61.055] [NeuralNetwork(12)] [warning] The issued warnings are orientative, based on optimal settings for a single network, if multiple networks are running in parallel the optimal settings may vary
Pipeline started - USB speed: SUPER
/Volumes/GoogleDrive-102536951206467607597/My Drive/Projects/Punchout/SkeletalTracker/depthai_blazepose/mediapipe_utils.py:246: RuntimeWarning: overflow encountered in exp
scores = 1 / (1 + np.exp(-scores))
Traceback (most recent call last):
File "/Volumes/GoogleDrive-102536951206467607597/My Drive/Projects/Punchout/SkeletalTracker/depthai_blazepose/demo.py", line 68, in
frame = renderer.draw(frame, body)
File "/Volumes/GoogleDrive-102536951206467607597/My Drive/Projects/Punchout/SkeletalTracker/depthai_blazepose/BlazeposeRenderer.py", line 158, in draw
self.draw_landmarks(body)
File "/Volumes/GoogleDrive-102536951206467607597/My Drive/Projects/Punchout/SkeletalTracker/depthai_blazepose/BlazeposeRenderer.py", line 114, in draw_landmarks
cv2.rectangle(self.frame, body.xyz_zone[0:2], body.xyz_zone[2:4], (180,0,180), 2)
TypeError: Argument 'thickness' is required to be an integer

Hello!

When running in xyz mode on an OAK D-Lite, the RGB image is out of focus, as it is set manually during initialization in https://github.com/geaxgx/depthai_blazepose/blob/main/BlazeposeDepthaiEdge.py#L238.

The comment mentions a calibration process - is this something I can replicate myself? I couldn't find any information on it in this repo.

Thank you, and sorry if I'm missing something obvious :)

I know this might be a question for the depthai github account but their examples do not seem to use your code structure.

I have multiple OAK-D pros connected to the same network, each with their own fixed IP address.
I am hoping to define in the code which device to connect to.

But doing this does not work:

        device_info = dai.DeviceInfo("10.100.0.21")
        self.device = dai.Device(device_info)

https://github.com/geaxgx/depthai_blazepose/blob/main/BlazeposeDepthaiEdge.py#L104

But in their example they are using a structure that does not seem to match the way you write your code:

device_info = dai.DeviceInfo("169.254.1.222")
# device_info = depthai.DeviceInfo("14442C108144F1D000") # MXID
# device_info = depthai.DeviceInfo("3.3.3") # USB port name

with dai.Device(pipeline, device_info) as device:
    qRgb = device.getOutputQueue(name="rgb", maxSize=4, blocking=False)
    while True:
        cv2.imshow("rgb", qRgb.get().getCvFrame())
        if cv2.waitKey(1) == ord('q'):
            break
``
https://docs.luxonis.com/projects/hardware/en/latest/pages/guides/getting-started-with-poe.html#manually-specify-device-ip

Thanks again for any advice.

Hi, I've just have a question for the two parameters : lm_input_length & pd_input_length
You have addressed to them static value, but what does they mean ? Because I saw that letter you used those value to divide the landmarks resulted by the camera

Thanks in advance

Hi, this is really a great work. However, when I am using your codes, I notice that the overall pose estimation is excellent, but the estimated position of thumbs are not so precise, especially whem palms are upwards. Actually, the palms are still backwards in 3d pose estimation while in reality they are upwards. It seems like that the model cannot detect the real position of pinky, index and thumb. Have you solved this problem?

The paper says:"we align the person so that the point between the hips is loceated at the center of the square image passed as the neural network input". Can you tell me where this part of the code is?