# Best practices for accelerating Python image processing on PYNQ?

**URL:** <https://discuss.pynq.io/t/best-practices-for-accelerating-python-image-processing-on-pynq/8933>\
**Category:** Support\
**Created:** [December 18, 2025, 9:42pm UTC](https://discuss.pynq.io/t/best-practices-for-accelerating-python-image-processing-on-pynq/8933 "2025-12-18T21:42:26Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![aria\_22](https://avatars.discourse-cdn.com/v4/letter/a/da6949/32.png) [@aria\_22](https://discuss.pynq.io/u/aria_22)\
**Post date:** [December 18, 2025, 9:42pm UTC](https://discuss.pynq.io/t/best-practices-for-accelerating-python-image-processing-on-pynq/8933/1 "2025-12-18T21:42:26Z")

</div>

Hi everyone — I’m experimenting with image processing workflows on a PYNQ board and trying to figure out whether it’s more efficient to offload parts of the pipeline into programmable logic or handle it in Python/PL together. I’m particularly interested in real-time tasks like streaming or object detection, and I want to see what others have done for similar workloads. For context, I came across this ESP32-CAM object counting project that shows how an ESP32-CAM can stream images over Wi-Fi and feed them into a Python/OpenCV system for real-time processing: [https://www.theengineeringprojects.com/2025/03/object-counting-project-using-esp32-cam-and-opencv.html](https://www.theengineeringprojects.com/2025/03/object-counting-project-using-esp32-cam-and-opencv.html) I’ve also seen some Raspberry Pi image-processing threads and Arduino forum projects where people push camera frames to local servers or PC apps for analysis. On PYNQ, should I be using custom overlays and DMA to feed frames directly into hardware accelerators, or is leveraging Python with PL only for pre/post processing usually sufficient? Any workflow tips or examples would be super helpful!