Info -interesting stats

[dobkeratops]

For comparison “CIFAR10 = 10 categories x 6000 examples” 60000 images but just 32x32 images. CIFAR100 = 60000images divided between more classes (600 each)

im wondering how many of our categories we could train something decent on

Urban environments
Road = 4994
Pavement = 3547
“Road vehicle”=Car,truck,bus,van = 3932
Building = 2305

Person=2381
Man=1020, woman=1774, (I wasn’t sure if the existing logic combined these into a “person” search bus as man+woman > person, maybe not?
combined counts
Man,woman,child,boy,girl = 2885
Person,man,woman,child ,boy,girl= 6898
People with common states -(eg “man/walking”
{Man,woman ,person} x {walking,sitting,running,standing} = 2837
All common “person” annotations (gender,states) = 8097
(Few more states = reclining,sittingCrossLegged,excercising,playingGuitar,reading,sleeping,leaning etc)

head/man =2036
Head/woman=3685

Animals
Dog=1367. Head=708.
Cat=693 head= 427
Lemur =1099
red_panda=878
All “quadrupedal_mammal” (dog cat horse cow etc) = 4605
Individually most of those only have ~100 examples

ungrouped..
Head=817 (these will be a mix of human and animal)
Hand=311
Foot=54

total annotations =?

[bbernhard]

Very cool, thanks for sharing!

Would be really interesting to see whether we could get a decently trained model out of that data 🤔

I wasn’t sure if the existing logic combined these into a “person” search bus as man+woman > person, maybe not?

At the moment there's no label substitution performed. It's on my Todo list, but I haven't had time to look into that :) So all the numbers are "raw" numbers :)

btw: Here are also some global stats, in case you are interested:

https://imagemonkey.io/statistics/contributions

I am always blown away by looking at those graphs. It's even more impressive when you consider that probably > 95% of that data was contributed by you. WOW!