search

eembc / energyrunner

The EEMBC EnergyRunner application framework for the MLPerf Tiny benchmark.
14 stars 5 forks source link

Is `Ground Truth` in `datasets/vww01/y_labels.csv` correct? #25

Closed xlindo closed 2 years ago

xlindo commented 2 years ago

No problem.

Please delete this issue. @petertorelli

Hi, For example, the first two pictures 000000343218.bin and 000000332570.bin both are class 1 in y_labels.csv. But actually whether in val2017 or in val2014, 000000343218 is person while 000000332570 is no person. Is there something wrong with my understanding?

  • 000000343218 in val2017
  • 000000332570 in val2017
petertorelli commented 2 years ago

Interesting, thanks for filing this bug.

@cskiraly This looks like a question for the MLPerfTiny working group.

xlindo commented 2 years ago

Interesting, thanks for filing this bug.

@cskiraly This looks like a question for the MLPerfTiny working group.

It is my fault. The fingers or legs are also involved. So it is correct.

petertorelli commented 2 years ago

@xlindo I'll keep it in the "closed" archive because I think that is a helpful observation. I thought "person" meant "full canonical biped", it would not occur to me that hand = person. But perhaps if the person-detector is used for controlling a CNC emergency shutoff if a person enters the frame, then a hand would certainly be an important thing to classify!

Peter