Interactively Transferring CNN Patterns for Part Localization
This addresses the challenge of learning correct patterns from noisy signals with limited supervision in part localization tasks, though it is incremental as it builds on existing CNN knowledge with manual refinement.
The paper tackles the problem of part localization in one/multi-shot learning by proposing a method that transfers latent activation patterns from a pre-trained CNN to a new model via human interaction, achieving superior performance in experiments.
In the scenario of one/multi-shot learning, conventional end-to-end learning strategies without sufficient supervision are usually not powerful enough to learn correct patterns from noisy signals. Thus, given a CNN pre-trained for object classification, this paper proposes a method that first summarizes the knowledge hidden inside the CNN into a dictionary of latent activation patterns, and then builds a new model for part localization by manually assembling latent patterns related to the target part via human interactions. We use very few (e.g., three) annotations of a semantic object part to retrieve certain latent patterns from conv-layers to represent the target part. We then visualize these latent patterns and ask users to further remove incorrect patterns, in order to refine part representation. With the guidance of human interactions, our method exhibited superior performance of part localization in experiments.