A three-part pipeline (blank-free rendering, instruction-aware visual-token pruning, and per-input compression settings) cuts visual tokens for code-image MLLM inputs by up to 71% without losing accuracy.
Humaneval-v: Systematic evaluation of visual reasoning in large multimodal models for code generation,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
CodeShrink: Adaptive Visual Compression for Efficient Multimodal Code Understanding
A three-part pipeline (blank-free rendering, instruction-aware visual-token pruning, and per-input compression settings) cuts visual tokens for code-image MLLM inputs by up to 71% without losing accuracy.