I need binocular video for an image processing application, so it seemed logical to "stack" cameraR and cameraL into a single frame using videobox and videomixer and then pass that into my appsink for processing as a single buffer. Unfortunately performance is terrible with jerky video and occasional messages about dropped frames. Using Ubuntu 10.04 64-bit, all patches as of this morning and its "stock" gstreamer-0.10.28 setup on an AMD quad core processor with 8GB RAM. (I had tried the PPA version a while back, but it hosed my system, recovering was a PITA so I'm not amenable to trying it again!) Sample pipeline with gst-launch: gst-launch v4l2src device=/dev/video1 ! ffmpegcolorspace ! video/x-raw-yuv,format=\(fourcc\)AYUV,width=640,height=480 !\ videobox border-alpha=0 top=-480 bottom=-48 ! videomixer name=mix sink_0::alpha=1.0 sink_1::alpha=1.0 ! ffmpegcolorspace !\ xvimagesink v4l2src device=/dev/video0 ! ffmpegcolorspace ! video/x-raw-yuv,format=\(fourcc\)AYUV,width=640,height=480 !\ ffmpegcolorspace ! mix. But running two instances gives smooth video despite using significantly more CPU according to top: gst-launch v4l2src device=/dev/video1 ! ffmpegcolorspace ! video/x-raw-yuv,format=\(fourcc\)AYUV,width=640,height=480 !\ ffmpegcolorspace ! xvimagesink v4l2src device=/dev/video0 ! ffmpegcolorspace !\ video/x-raw-yuv,format=\(fourcc\)AYUV,width=640,height=480 ! ffmpegcolorspace ! xvimagesink I've 5 different v4l2 capture devices (two PCI and 3 USB) and all pairs have the perfromance problem including one USB device that has two full frame rate capture inputs (Sensoray 2255S). All pairings give smooth video with two instances. The video has perfect "genlock" as its one VCR output fed to the capture devices with a distribution amplifier. The 48 extra scan lines at the bottom of the stacked frames I planned to fill with barcoded metadata, removing them didn't change anything. In fact I can run both gst-launch commands from separate terminal windows and its very obvious when the videobox/videomixer drops frames. Is there something I'm missing that can fix this? The two cameras in a single buffer would seem to make the downstream code much more understandable, but I can use the two instance solution. Any rational reason why videobox supports video/x-raw-gray and videomixer does not?