Search papers, labs, and topics across Lattice.
3
0
4
1
Achieving 94.7% performance retention with just 128 tokens, 3DZip revolutionizes token compression for 3D question answering.
RestoreKV boosts performance by recovering lost context information, achieving up to 86.4% accuracy at 16脳 compression while incurring minimal overhead.
Diversity-based visual token pruning in LVLMs might be increasing hallucination, as it often fails to preserve feature diversity and struggles with simple images compared to attention-based methods.
