Image-level deep tags: essentially equivalent to the proto above but containing tags that are computed at the image level. These image signals are maintained by the Visual Shopping team (visual-shopping@). If you do use the signals, please add an entry in go/ShoppingImageAttributeClients to be notified for model upgrade. We recommend our clients against using the raw confidence value directly. Instead, the clients should use the library, cs/ads/shopping/visual/deeptags/public/single_tag.h and cs/ads/shopping/visual/deeptags/public/single_scored_tag.h to specify an operating point in terms of precision or recall. See the following code example: http://google3/shopping/visual/explore_looks/looks_offline_pipeline.cc?l=268&rcl=304165166 model_outputs is a repeated field. Please check version to get the model you desire to use, instead of indexing the model_outputs directly e.g. model_outputs(0). We will remove the old versions in the future and this will lead to incorrect model. Models: As of Q2 2020, we have two models running within Shopping: model one only has the overlay tag, which we are deprecating, and model two has the tags specified in go/VisualShoppingImageAttributes.

Attributes

• modelOutputs (type: list(GoogleApi.ContentWarehouse.V1.Model.CommerceDatastoreImageDeepTagsModelOutput.t), default: nil) - The set of outputs for a series of model versions. The size of this field should not extend beyond 4 at any time: two versions for slow-update track dependencies, and two versions for fast-update track dependencies.