CLAVER adds a cross-frame temporal attention mask (Kronecker mask) and LLM-generated interpretive action prompts to CLIP, improving video action recognition.
The low-rank bottleneck is vital, because it may lead to that many rows of the attention map are seriously homogenized
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Kronecker Mask and Interpretive Prompts are Language-Action Video Learners
CLAVER adds a cross-frame temporal attention mask (Kronecker mask) and LLM-generated interpretive action prompts to CLIP, improving video action recognition.