luxonis / depthai

DepthAI Python API utilities, examples, and tutorials.
https://docs.luxonis.com
MIT License
938 stars 232 forks source link

[BUG] {Image artifacts when using onboard encoding on FFC-6P and RCV3} #1105

Open JordanGell opened 1 year ago

JordanGell commented 1 year ago

Dear Luxonis, we've been having the following issue with the FFC-6P

Hardware

-OAK FFC-6P -socket CAM_B : OV9282 640 x 480 focus:fixed - MONO -socket CAM_C : OV9282 640 x 480 focus:fixed - MONO -socket CAM_F : IMX412 4056 x 3040 focus:fixed - COLOR

Software -DepthAI version: 2.22.0.0 -Branch RCV3_Develop

Description

When using the onboard video encoding on the FFC-6P we get good frames with a stationary subject however with any motion of the subject artefacts appear (Example video attached). We are also unable to get video from the IR cameras when using the onboard video encoding, only empty files.

Expected behavior

Frames encoded on the FFC-6P for all three cameras without artefacts

Attach system log log_system_information.txt

Video https://github.com/luxonis/depthai/assets/104423166/ecf4a90b-4a2e-472c-8eb0-c8da503931d8

brmarkus commented 1 year ago

Can you maybe take one of the encoded (non-empty, non-black) frames and highlight artefacts, please? It might be my video-player, or I just missed the important frames - I only see "noise" due to the encoder using a low bitrate (and/or operating in low-latency mode).

Example: image

JordanGell commented 1 year ago

We use the auto bitrate setting, with the latest 6P firmware, and we do not turn on low-latency (what is this setting?) there are a couple of bits that look like artifcacts but mainly it's the low quality output or noise that we're concerned with. We do not get this behaviour when not using the Video Encoder. This is using the H264 presets - when using the HEVC, the quality is worse and when using MJPEG we recieve one (clear good quality) frame then the entire pipeline locks up.