Search papers, labs, and topics across Lattice.
Sun Yat-sen University
2
0
4
Fewer than two edits can enable 76.2% of unsafe images to bypass safety classifiers while retaining their malicious intent, exposing critical vulnerabilities in current moderation systems.
Structured supervision can enable 3B VLMs to outperform 32B VLMs on dense-scene reasoning tasks, suggesting a path to efficient and reliable visual understanding.
