Text-to-Image Generation on COCO-CN ZH (test)
14.35FIDGround-Truth EN Captions
Evaluation Results
| Method | Links | |
|---|---|---|
| Ground-Truth EN CaptionsSource=Human-written2023.05 | 14.35 | |
| ENSADGenerator=Frozen2023.05 | 14.52 | |
| ENSAD + Fine-Tune GGenerator=Fine-tuned, Strategy=Mean Pooling2023.05 | 15.14 | |
| ZERO-SHOT TRANSFERFine-tuning=false2023.05 | 15.57 | |
| Mean PoolingStrategy=Mean Pooling2023.05 | 16.47 |