Concept早期融合 / Early Fusion
Feature-level Fusion
A fusion paradigm that concatenates or aggregates feature vectors from multiple sources after feature extraction and before similarity computation. It retains rich original information but requires good feature alignment and is sensitive to noise and missing modalities.
Timeline (last 90 days)
Oct 30
Feature-level fusion (early fusion) aggregates features from multiple sources after feature extraction and before similarity computation, preserving more detail but being sensitive to noise; score-level fusion (late fusion) first computes similarities independently and then merges them, being more robust but potentially losing cross-view association information.
Unverified50%