Writing a simple vAccel Python application🔗
For a simple example of using the vaccel package, you can replicate in Python the image classification example from Running the examples.
Create a new Python file called classify.py with the following content:
classify.py
#!/usr/bin/python3
import sys
from pathlib import Path
from vaccel.session import Session
def main():
with Path(sys.argv[1]).open("rb") as f:
image = f.read()
sess = Session()
print(f"Session id is {sess.id}")
res = sess.classify(image)
print(res)
if __name__ == "__main__":
main()
As with the original classify example, to execute classify.py you need to configure vAccel:
Assuming vAccel in installed at /usr/local, running the Python classification should give you:
$ python3 classify.py /usr/local/share/vaccel/images/example.jpg
Session id is 1
('This is a dummy classification tag!', 'This is a dummy imgname!')
By adding debug level logging:
you can get the verbose version of the output:
$ python3 classify.py /usr/local/share/vaccel/images/example.jpg
2025.04.13-21:21:39.87 - <debug> Initializing vAccel
2025.04.13-21:21:39.87 - <info> vAccel 0.6.1-194-19056528
2025.04.13-21:21:39.87 - <debug> Config:
2025.04.13-21:21:39.87 - <debug> plugins = libvaccel-noop.so
2025.04.13-21:21:39.87 - <debug> log_level = debug
2025.04.13-21:21:39.87 - <debug> log_file = (null)
2025.04.13-21:21:39.87 - <debug> profiling_enabled = false
2025.04.13-21:21:39.87 - <debug> version_ignore = false
2025.04.13-21:21:39.87 - <debug> Created top-level rundir: /run/user/1002/vaccel/WTLhQW
2025.04.13-21:21:39.87 - <info> Registered plugin noop 0.6.1-194-19056528
2025.04.13-21:21:39.87 - <debug> Registered op noop from plugin noop
2025.04.13-21:21:39.87 - <debug> Registered op blas_sgemm from plugin noop
2025.04.13-21:21:39.87 - <debug> Registered op image_classify from plugin noop
2025.04.13-21:21:39.87 - <debug> Registered op image_detect from plugin noop
2025.04.13-21:21:39.87 - <debug> Registered op image_segment from plugin noop
2025.04.13-21:21:39.87 - <debug> Registered op image_pose from plugin noop
2025.04.13-21:21:39.87 - <debug> Registered op image_depth from plugin noop
2025.04.13-21:21:39.87 - <debug> Registered op exec from plugin noop
2025.04.13-21:21:39.87 - <debug> Registered op tf_session_load from plugin noop
2025.04.13-21:21:39.87 - <debug> Registered op tf_session_run from plugin noop
2025.04.13-21:21:39.87 - <debug> Registered op tf_session_delete from plugin noop
2025.04.13-21:21:39.87 - <debug> Registered op minmax from plugin noop
2025.04.13-21:21:39.87 - <debug> Registered op fpga_arraycopy from plugin noop
2025.04.13-21:21:39.87 - <debug> Registered op fpga_vectoradd from plugin noop
2025.04.13-21:21:39.87 - <debug> Registered op fpga_parallel from plugin noop
2025.04.13-21:21:39.87 - <debug> Registered op fpga_mmult from plugin noop
2025.04.13-21:21:39.87 - <debug> Registered op exec_with_resource from plugin noop
2025.04.13-21:21:39.87 - <debug> Registered op torch_jitload_forward from plugin noop
2025.04.13-21:21:39.87 - <debug> Registered op torch_sgemm from plugin noop
2025.04.13-21:21:39.87 - <debug> Registered op opencv from plugin noop
2025.04.13-21:21:39.87 - <debug> Registered op tflite_session_load from plugin noop
2025.04.13-21:21:39.87 - <debug> Registered op tflite_session_run from plugin noop
2025.04.13-21:21:39.87 - <debug> Registered op tflite_session_delete from plugin noop
2025.04.13-21:21:39.87 - <debug> Loaded plugin noop from libvaccel-noop.so
2025.04.13-21:21:39.88 - <debug> New rundir for session 1: /run/user/1002/vaccel/WTLhQW/session.1
2025.04.13-21:21:39.88 - <debug> Initialized session 1
Session id is 1
2025.04.13-21:21:39.88 - <debug> session:1 Looking for plugin implementing VACCEL_OP_IMAGE_CLASSIFY
2025.04.13-21:21:39.88 - <debug> Returning func from hint plugin noop
2025.04.13-21:21:39.88 - <debug> Found implementation in noop plugin
2025.04.13-21:21:39.88 - <debug> [noop] Calling Image classification for session 1
2025.04.13-21:21:39.88 - <debug> [noop] Dumping arguments for Image classification:
2025.04.13-21:21:39.88 - <debug> [noop] model: (null)
2025.04.13-21:21:39.88 - <debug> [noop] len_img: 79281
2025.04.13-21:21:39.88 - <debug> [noop] len_out_text: 500
2025.04.13-21:21:39.88 - <debug> [noop] len_out_imgname: 500
2025.04.13-21:21:39.88 - <debug> [noop] will return a dummy result
2025.04.13-21:21:39.88 - <debug> [noop] will return a dummy result
('This is a dummy classification tag!', 'This is a dummy imgname!')
2025.04.13-21:21:39.88 - <debug> Released session 1
2025.04.13-21:21:39.89 - <debug> Cleaning up vAccel
2025.04.13-21:21:39.89 - <debug> Cleaning up sessions
2025.04.13-21:21:39.89 - <debug> Cleaning up resources
2025.04.13-21:21:39.89 - <debug> Cleaning up plugins
2025.04.13-21:21:39.89 - <debug> Unregistered plugin noop
The vAccel output of classify and classify.py should be almost identical.
Info
For a full example with near identical functionality to the C classify you can look at the image classification example of the vaccel-python repository,