What is BSafe?

: It is a tool to analyze videos and images for Brand Safety concerns.

What does BSafe offer to help you with ?

: Index videos automatically, at scale, for concepts involving violence, adult content, etc.

Computes unique fingerprints for videos and images, so that a content-based blacklist can be maintained independent of URLs.
#Test
##Test
###Test
####Test
#Test 3

• item1
• item 2
  – item3
  There are three XenTag Standard Models.

| Models/Desc | Name | Details | OutputValues | Model Number |
|------------- |------------------- |------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |-------------------------------------------------- |-------------- |
| Library | caffe-ilvsrc12-nc | Reference model from ImageNet benchmarks competition. This is a non-commercial version of the ilsvrc12 top-five Convolutional Neural Net mode, trained on 1.4M images, which recognizes objects from 1K categories from Imagenet. Each category was trained on ~1.5k images. | 1K different Imagenet nouns, with probabilities. | 1 |
| Multi-class | CatsVsDogs | This model was trained, as a multi-Class linear model, where the feature detector is a piece of a Library grade model/ This model fpllows the Kaggle DogsVCats competition, and is based on “nolearn” code from D. Nouri | 1=Dog, 0=Cat | 2 |
| Concept | Sports | This model is based on a bag of WordNet words defining a Concept(e.g. Sports) The Concept has a tree span in in WordNet, and is intersected with the Wordnet tree span of the underlying Library model categories. If the Library outputs form a probability distribution, then the Pr[Concept]===Pr[Concept ^ LibOutputs] | | 4 |