Collected sources and patterns will appear here. Add from search, explore, or the patterns library.
Tensor<EncoderFeatures> -> Tensor<ObjectQueries>
Select the top-K spatial features from an encoder to initialize decoder object queries by scoring each feature pixel using a joint metric of classification confidence and localization bounding-box quality.
Problem it solves
Using classification score alone to select object query priors selects queries with high classification confidence but poor localization accuracy.
Consumes
Emits
The real projects this mechanism was found in. Attribution is the point — this is how the best teams actually do it.