fang196

The code has been released in #53 .

Same problem when zero-shot prompt-tuning on lvis dataset. ``` 04/03 23:59:11 - mmengine - INFO - Epoch(test) [6603/6603] lvis/bbox_AP: 0.0000 lvis/bbox_AP50: 0.0010 lvis/bbox_AP75: 0.0000 lvis/bbox_APs: 0.0000 lvis/bbox_APm: 0.0000 lvis/bbox_APl: 0.0000...

Solve this problem. 1. Using the clip large model to extract text embeddings. 2. Using yolo world v2 model as pretrained model, for example, yolo_world_v2_l_clip_large_o365v1_goldg_pretrain-8ff2e744.pth. ``` mmengine - INFO -...