Running Google Mediapipe Hand Tracking models on DepthAI hardware (OAK-1, OAK-D, ...)

There is a version for OpenVINO there : openvino_hand_tracker

Install DepthAI gen2 python package:

python3 -m pip install -r requirements.txt

To use the color camera as input :

python3 HandTracker.py

To use a file (video or image) as input :

python3 HandTracker.py -i filename

To enable gesture recognition :

python3 HandTracker.py -g

To run only the palm detection model (without hand landmarks), use --no_lm argument. Of course, gesture recognition is not possible in this mode.

Use keypress between 1 and 7 to enable/disable the display of hand features (palm bounding box, palm landmarks, hand landmarks, handedness, gesture,...)

You can find the models palm_detector.blob and hand_landmark.blob under the 'models' directory, but below I describe how to get the files.

1. Clone this github repository in a local directory (DEST_DIR)
2. In DEST_DIR/models directory, download the source tflite models from Mediapipe:

• Palm Detection model
• Hand Landmarks model

3. Install the amazing PINTO's tflite2tensorflow tool. Use the docker installation which includes many packages including a recent version of Openvino.
4. From DEST_DIR, run the tflite2tensorflow container: ./docker_tflite2tensorflow.sh
5. From the running container:

cd resources/models
./convert_models.sh

The convert_models.sh converts the tflite models in tensorflow (.pb), then converts the pb file into Openvino IR format (.xml and .bin), and finally converts the IR files in MyriadX format (.blob).

By default, the number of SHAVES associated with the blob files is 4. In case you want to generate new blobs with different number of shaves, you can use the script gen_blob_shave.sh:

# Example: to generate blobs for 6 shaves
./gen_blob_shave.sh -m pd -n 6   # will generate palm_detection_sh6.blob
./gen_blob_shave.sh -m lm -n 6   # will generate hand_landmark_sh6.blob

Explanation about the Model Optimizer params :

• The preview of the OAK-* color camera outputs BGR [0, 255] frames . The original tflite palm detection model is expecting RGB [-1, 1] frames. --reverse_input_channels converts BGR to RGB. --mean_values [127.5,127.5,127.5] --scale_values [127.5,127.5,127.5] normalizes the frames between [-1, 1].
• The images which are fed to hand landmark model are built on the host in a format similar to the OAK-* cameras (BGR [0, 255]). The original hand landmark model is expecting RGB [0, 1] frames. Therefore, the following arguments are used --reverse_input_channels --scale_values [255.0, 255.0, 255.0]

Blob models vs tflite models The palm detection blob does not exactly give the same results as the tflite version, because the tflite ResizeBilinear instruction is converted into IR Interpolate-1. Yet the difference is almost imperceptible thanks to the great help of PINTO (see issue ).