Closed AditiBhiwaniwala closed 7 years ago
@AditiBhiwaniwala can you attach the original image? Please note that the fork of darknet we are using is not on par with the current master branch of original darknet. I.e. the latest version of the darknet may produce better/different results.
@OrKoN I can't confirm this if I use the image with configs in the /test
folder ... I get this classifications (w/o the Parking Meter category):
[ { x: 0.7554311156272888,
y: 0.055360980331897736,
w: 0.07247214019298553,
h: 0.0692552700638771,
prob: 0.378817081451416,
name: 'car' },
{ x: 0.7900861501693726,
y: 0.041906263679265976,
w: 0.04912581667304039,
h: 0.05180671066045761,
prob: 0.2652934193611145,
name: 'car' },
{ x: 0.03810616955161095,
y: 0.17536404728889465,
w: 0.0602089948952198,
h: 0.32217973470687866,
prob: 0.42542749643325806,
name: 'person' },
{ x: 0.08943124115467072,
y: 0.16063717007637024,
w: 0.11497301608324051,
h: 0.3066703975200653,
prob: 0.6008384227752686,
name: 'person' },
{ x: 0.4950035512447357,
y: 0.43869340419769287,
w: 0.7062614560127258,
h: 1.00821053981781,
prob: 0.7091837525367737,
name: 'car' } ]
I am attaching an image in which a car is detected as a parking meter whereas when using the original darknet library the result is correct. If i am using same cfg file, weight file and data file then why is this error coming.