Hey, your demo video seems to be very smooth. Did you actually use a GPU for producing the demo? I tried to run the script on my laptop with i7 but without GPU and got ~2-3 FPS.

Hi I have tried the below code to draw bbox around hand but not drawing properly

(handtracker.py)

bounding box offsets, width and height

    dx, dy, w, h = candidate_detect[box_ids, :4]

bbox_dict = {
"x": dx,
"y": dy,
"w": w,
"h": h
}
cv2.rectangle(frame, (bbox_dict['x'], bbox_dict['y']),
(bbox_dict['x'] + bbox_dict['w'], bbox_dict['y'] + bbox_dict['h']), (0, 255, 0), 1)

please advise

Hi,

you can please explain to did you created the withoud custom op version of the mediapipe model?
i need other model without custom op.

thanks :)

How do I detect both hands? I've already created two connections to recognize both hands, but it didn't work.

I implement your python code and got keypoint value but I can't got x, y coordinates normalization, So How to improve coordinates normalization form your code ?

Thank you so much.

Hi,thanks for you share about hand_tracking ,it's a nice job.
I want to consult that how could we detect multi hands and these points?
Thx!

Hi metalwhale,

First of all, nice job. Thank you for trasnfering the google code. Right now I have the problem that it can't detection the hand. I was guessing if it is because the background is too noisy but it still didn't work even I applied the clear and white background. I firstly modified code to run with Kinect but it still didn't work with webcam. Could you help me with it? Thank you so much.

How i can use this git for hand gusture recognition?
I would appreciate for your help, thanks.

Thank you for share this project.
Input same(similar) image to google and your project respectly, however get large distinguish between output keypoints. Have you find this question?
A simple test, download google app and install, open google app and your project sync and than compare result.

Congratulation for your awesome work!!
When I try to run the file, it appears this error:

File "c:\ User_name \python tutorial\raspberrypicode\palmgit\hand_tracking-master2\run.py", line 67, in
detector = HandTracker(
File "c:\ User_name \python tutorial\raspberrypicode\palmgit\hand_tracking-master2\src\hand_tracker.py", line 32, in init
self.interp_palm = tf.lite.Interpreter(palm_model)
File "C:\Users\ User_name \AppData\Local\Programs\Python\Python310\lib\site-packages\tensorflow\lite\python\interpreter.py", line 455, in init
_interpreter_wrapper.CreateWrapperFromFile(
ValueError: Could not open 'models/palm_detection_without_custom_op.tflite'.

Any suggestions ?

The frame rate seems to be extremely slow. As I move my hand across the camera, the display updates not in a smooth manner but very shakily (seems to be low frame rate?) The result is nothing like the once seem in the project description.

Hi, thank you for sharing excellent project. I have been trying to change to the threshold (even to 0.95) to eliminate false detection, but it does not help.

Do you have any suggestion?

@metalwhale

thanks for amazing work.

any plan to add/support for the gesture recognition (palm open, close)?

or if you know any repo please advise here.,
thanks in advance

thanks for your nice work.
please can you write a README about how to generate this file
THANK YOU AGAIN ~

Hi, @metalwhale ,its me again ^-^.
I test your work, and find that :
1.The hand key points that be detected seem not steady like your result, they always shake.
2.Sometimes it detect my face as a hand by mistake !
3. How can I find the hand key points' confidence?
Could you have some advise about that? Thanks alot.