Norm-guided Adaptive Visual Embedding for Zero-Shot Sketch-Based Image Retrieval.
-Siamese network, pretrained SeResNet, transformation layers ( linear layers )
-Training loop using mutli task learning by "Multi-Task Learning Using Uncertainty to Weigh Losses for Scene Geometry and Semantics" paper in order not to noise regularizer.
- Test loop (It is completely different from training loop)