PaCE: Unified Multi-modal Dialogue Pre-training with Progressive and Compositional Experts

Benchmark Model Rank Results
dialogue-state-tracking-on-simmc2-0PaCE#1Act F1: 97.1Slot F1: 87.0
image-retrieval-on-photochatPaCE#1R1: 15.2R@10: 49.6R@5: 36.7Sum(R@1,5,10): 101.5
multimodal-intent-recognition-on-mmdialogPaCE#1F1: 77.6
multimodal-intent-recognition-on-photochatPaCE#1F1: 63.8Precision: 63.3Recall: 68
response-generation-on-simmc2-0PaCE#1BLEU: 34.1
text-retrieval-on-image-chatPaCE#1R@1: 51.9R@5: 76.8Sum(R@1,5): 128.7