Text-to-audio retrieval on iNatSounds 2024 (val)
50.05mAP@1000 (Amphi)BioLingual with 256-bit Hashing
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| BioLingual with 256-bit HashingHashing Bits=256-bit, Feature Representation=Continuous, Training Strategy=LoRA Fine-tuned2026.01 | 50.05 | 58.21 | 50.3 | 59.28 | 94.64 | 52.08 | |
| BioLingual with 256-bit HashingHashing Bits=256-bit, Feature Representation=Binary (Hashing)2026.01 | 48.49 | 55.87 | 49.9 | 46.92 | 90.51 | 48.61 | |
| BioLingual with 128-bit HashingHashing Bits=128-bit, Feature Representation=Continuous, Training Strategy=LoRA Fine-tuned2026.01 | 46.51 | 55.65 | 44.21 | 54.11 | 96.29 | 49.46 | |
| BioLingual with 128-bit HashingHashing Bits=128-bit, Feature Representation=Binary (Hashing)2026.01 | 44.13 | 52.98 | 39.45 | 42.63 | 88.35 | 44.59 | |
| BioLingualFeature Representation=Original Embedding2026.01 | 39.06 | 48.54 | 37.84 | 51.59 | 93.11 | 45.02 |