Announcement

Collapse
No announcement yet.

TensorFlow giving the wrong result

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • TensorFlow giving the wrong result

    hello,

    has as anyone been using TensorFlow? We are using the standard program for testing. I has the kids run the test 15 times with the gold mineral in various locations. 11 of the 15 it chose left and was correct only 27% of the time. Looking at the programing it appears to start left so I believe the issue is in the programming. It is like the program latches on left and cannons get off it.

    when we re-ran some tests we notices the gold and silver were CORRECTLY labeled, how can that be, yet it will provide the wrong location

    I read the instruction manual and it

  • #2
    Sorry last bit got cut off. The instruction manual mentions the confidence level. Where is this to be inserted or modified? The details are not very specific, create an object? In what? Where is the 40% confidence located.

    Comment


    • #3
      We are attempting to use TensorFlow and have randomly seen the same issue you are seeing , but our consistently bad experience has more to do with sensing the minerals. It will not reliably detect all three at the same time and is very lighting dependent. It was a disaster at our last competition. Worked pretty well on the test field, then didn't on the competition field as the lighting was different.

      What program are you using? To answer your question, object tracking and confidence are located in the initialization section. Note there is an additional issue where the bot cannot sit in initialization too long or it will lockup.

      Comment


      • #4
        What do you mean by lock up?

        Comment


        • #5
          It would lose connectivity, so it needed a reboot. In our last competition, it happened three times, due to other teams delaying start. I read of a workaround by calling telemetry update during the Wait for Start loop, but we have elected to simply initializing TensorFlow after start.

          Comment


          • #6
            Which phones you are using if it e4 for DS then it will disconnect in initialization after 20-30 secs. There is another thread on this

            Comment


            • #7
              We are using Moto G 2nd gen phones

              Comment


              • #8
                Can you try just sample Tensorflow code? You do not need exp hub , Gamepad or battery or anything. See if it works reliably. Of it works then go ahead and modify sample to detect only one or two mineral at a time

                Comment


                • #9
                  We have been doing a lot of testing, and I am pretty confident that the models supplied with the code are no good. I have a theory what they did wrong in building the models, but will hold off sharing until I prove my theory. We have a workaround now and it seems to work 100%, but you cannot use the sample code as supplied as it will not reliably detect three minerals at a time due to the model issue.

                  Comment


                  • #10

                    For reliable results from the TensorFlow Example OpMode (in Blocks), make sure the phone is set to Auto-rotate.

                    This is clearly explained here:
                    https://github.com/ftctechnh/ftc_app...ge-orientation

                    The overall OpMode documentation is here:
                    https://github.com/ftctechnh/ftc_app...ection-Op-Mode


                    Without this, your tests so far have displayed the Gold mineral's relative vertical position in the image, rather than its relative horizontal position. This would randomly seem 'right' or 'wrong', creating much frustration. Been there!

                    If for some reason you want the phone locked to Portrait mode, in the Example code (in Blocks) simply change the 3 instances of Recognition.Left to Recognition.Top

                    I will post this note in the several threads reporting unreliable TensorFlow results.

                    Comment

                    Working...
                    X