The authors introduce SemanticMap, a cosine-similarity based evaluation metric for cross-modal retrieval, and a single-stream network that encodes text as images, but the metric can be trivially gamed by collapsing embeddings and the model underperforms on standard R@K.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2019 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Do Cross Modal Systems Leverage Semantic Relationships?
The authors introduce SemanticMap, a cosine-similarity based evaluation metric for cross-modal retrieval, and a single-stream network that encodes text as images, but the metric can be trivially gamed by collapsing embeddings and the model underperforms on standard R@K.