Search papers, labs, and topics across Lattice.
Affiliation:
3
0
3
Standard open-vocabulary detectors fail on nearly 70% of ambiguous, task-driven queries, but layering structured semantic retrieval and dynamic LLM concept expansion over frozen detectors pushes grounding success to 85%.
By encoding anatomical contours as learnable query embeddings, CGQR-Net achieves state-of-the-art cardiac ultrasound segmentation by explicitly incorporating structural priors, overcoming limitations of purely appearance-based methods.
Finally, a way to measure how efficiently a sketch conveys meaning, moving beyond simple recognition accuracy.
