How to Maintain High Resolution and FPS

kavya

Hi everyone,

I’m running a DepthAI pipeline on my OAK-D LR camera, and I’ve noticed a big drop in performance when connecting via PoE compared to USB-C.

Setup:
- OAK-D LR connected via PoE (48V, 15.4W injector — standard PoE, not PoE+)
- Running a custom DepthAI + YOLOv8 pose estimation pipeline
- Depth and RGB streams enabled
- Previously used 640×480 resolution for the camera nodes
- Tested with 320×240, and FPS improved significantly
Issue:
When connected via USB-C (with lower resolution), the pipeline runs smoothly at a good frame rate.
However, when using PoE, the FPS drops to around 1 FPS, even with the same pipeline and model.
Observations:
- Reducing the resolution to 320×240 increases FPS noticeably.
Question:
What are the best practices or recommended settings to:
- Maintain higher FPS while keeping 640×480 (or higher) resolution, and
- Optimize PoE performance for compute-heavy pipelines like YOLOv8 pose detection?
  
  i've went throught the documents for optimizing fps .
  here is the latency
  
  and bandwidth
  
  and
  System Performance Metrics
- Memory Usage: ~135/333 MiB DDR, CMX almost full (2.48/2.5 MiB)
- CPU Usage:
  - LeonOS: 50–90%
  - LeonRT: 1–6%

OskarSonc

Hey @kavya

Looks like your PoE link is limited to 100 Mb/s — the 69 Mb/s result, which is low. That caps raw RGB/depth to ~1 FPS.

Try with a Gigabit PoE(+) switch/injector and CAT5e/6 cable — bandwidth should jump >200 Mb/s.
Also avoid sending raw frames: run NN on-device, send only detections or encoded (H.265/H.264) video, and limit XLink FPS for previews.

Thanks,
Oskar

kavya

OskarSonc I’ve upgraded to a CAT6 cable and a Wi-Fi 7 router (Archer BE400) — it’s working perfectly now. Thanks a lot!