Sorry about that. I broke the example. I've updated it and it should work now.
https://github.com/beagleboard/cloud9-examples/commit/210388017fcb233c2f422d54af293bb8d5c94bc2 I was visiting the TI office and talking to the developers about the performance of this example. According to profiles, it should run up to 60fps. I attempted to make some changes to speed it up, but I did it wrong. You can group different layers in the network to run on different processors. For this classifier network, it is said to be fastest to run the first 11 stages on EVEs as fixed-point processes and then run the last 3 layers as floating-point processes on the C66 DSPs. And, because we'd only be running 3 layers on the DSPs, we only need a single DSP. Anyway, I didn't assign the layers properly and I still need to look at the code a bit more to set them properly. For now, I've just switched back to running on all 14 layers on 4 EVEs. The 30fps data from the camera seems to be reasonably processed with this configuration. I picked up a Logitech C922 that is capable of doing 60fps and I'll be looking to update the demo to run that way soon and finishing up the segmentation demo. Checking the commit-log is a nice way to check-up on me, even if my comments aren't the best. The errors are mostly due to the fact that I'm learning as well. I'm trying to get the TI developers to use my methodology of single-file mjpg-streamer filters rather then OpenCV desktop apps as I feel those the desktop apps are overly complex and don't represent an embedded developer's use-case. They are pretty reasonably documented, but, as you can see, it is taking me some time to understand them. Some additional debug visibility needs to be added to my approach and I'll be chatting to the TI developers about that some in my call later today about this stuff. Development work is on-going for Tensorflow Lite support. All should be much easier once that lands. And, yes, I keep talking about TI as if I don't work there, and I do work there, but my working with open source developers all day keeps me from adopting certain development processes other TI developers take as granted. I don't install Code Composer Studio. I don't setup an Open Embedded build environment. I don't cross-compile. I don't setup JTAG. I hope you get the idea. On Wednesday, October 23, 2019 at 12:46:26 AM UTC-4, Jon Morss wrote: > > > Yeah, I always find it suspect when am example is posted and demo'd but > does not seem to work for others. > > Headbanging continues. > > Jon > > On Tuesday, October 22, 2019 at 4:03:38 PM UTC-7, Dobrin Alexiev wrote: >> >> In my case I also see often ping-pong_ball, or more often "segmentation >> fault". >> I wonder how can I debug this? >> >> >> On Sunday, October 20, 2019 at 2:11:54 AM UTC-4, Jon Morss wrote: >>> >>> I am attempting to run the TIDL example with a Beaglebone AI and the >>> only thing it seems to report identifying is a ping-pong, although I am not >>> presenting a ping pong to the camera. I am using a Logitech C920 camera >>> and have performed all of the updates to the system, so am not sure what I >>> am missing. >>> >>> This is what I see when running the classification.tidl.cpp example: >>> >>> sudo mjpg_streamer -i "input_opencv.so -r 640x480 --filter ./ >>> classification.tidl.so" -o "output_http.so -p 8080 -w >>> /usr/share/mjpg-streamer/www" >>> [sudo] password for debian: >>> MJPG Streamer Version.: 2.0 >>> i: device........... : default >>> i: Desired Resolution: 640 x 480 >>> i: filter........... : ./classification.tidl.so >>> i: filter args ..... : >>> Initializing filter >>> loading configuration >>> allocating execution object pipelines (EOP) >>> allocating executors >>> allocating individual EOPs >>> allocating I/O memory for each EOP >>> Allocating input and output buffers >>> Allocating input and output buffers >>> Allocating input and output buffers >>> Allocating input and output buffers >>> num_eops=4 >>> About to start ProcessFrame loop!! >>> http://localhost:8080/?action=stream >>> o: www-folder-path......: /usr/share/mjpg-streamer/www/ >>> o: HTTP TCP port........: 8080 >>> o: HTTP Listen Address..: (null) >>> o: username:password....: disabled >>> o: commands.............: enabled >>> (722)=ping-pong_ball >>> (722)=ping-pong_ball >>> (722)=ping-pong_ball >>> (722)=ping-pong_ball >>> (722)=ping-pong_ball >>> (722)=ping-pong_ball >>> >>> >>> This is what I see from dmesg: >>> >>> [20753.769040] usb 1-1: New USB device found, idVendor=046d, idProduct= >>> 082d >>> [20753.769075] usb 1-1: New USB device strings: Mfr=0, Product=2, >>> SerialNumber=1 >>> [20753.769097] usb 1-1: Product: HD Pro Webcam C920 >>> [20753.769118] usb 1-1: SerialNumber: C0DB0F6F >>> [20754.099831] uvcvideo: Found UVC 1.00 device HD Pro Webcam C920 (046d: >>> 082d) >>> [20754.120146] uvcvideo 1-1:1.0: Entity type for entity Processing 3 >>> was not initialized! >>> [20754.120179] uvcvideo 1-1:1.0: Entity type for entity Extension 6 was >>> not initialized! >>> >>> [20754.120323] uvcvideo 1-1:1.0: Entity type for entity Extension 11 >>> was not initialized! >>> [20754.125089] input: HD Pro Webcam C920 as /devices/platform/ >>> 44000000.ocp/488c0000.omap_dwc3_2/488d0000.usb/xhci-hcd.1.auto/usb1/1-1/ >>> 1-1:1.0/input/input3 >>> [20754.135851] usbcore: registered new interface driver uvcvideo >>> [20754.135871] USB Video Class driver (1.1.1) >>> [20754.437849] usbcore: registered new interface driver snd-usb-audio >>> [20867.134498] usb 1-1: reset high-speed USB device number 3 using xhci- >>> hcd >>> [20867.558788] omap-iommu 58882000.mmu: 58882000.mmu: version 2.1 >>> [20867.605127] omap_hwmod: mmu0_dsp2: _wait_target_disable failed >>> [20867.605206] omap-iommu 41501000.mmu: 41501000.mmu: version 3.0 >>> [20867.605483] omap-iommu 41502000.mmu: 41502000.mmu: version 3.0 >>> [20867.619103] omap_hwmod: mmu0_dsp1: _wait_target_disable failed >>> >>> >>> Am I missing a step? >>> >>> Cheers, >>> >>> Jon >>> >> -- For more options, visit http://beagleboard.org/discuss --- You received this message because you are subscribed to the Google Groups "BeagleBoard" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/beagleboard/94498dec-c8ec-4798-ae93-2b843980fb08%40googlegroups.com.
