FoundationVision / GLEE

[CVPR2024 Highlight]GLEE: General Object Foundation Model for Images and Videos at Scale
https://glee-vision.github.io/
MIT License
1.02k stars 82 forks source link

关于coco指标复现的问题 #9

Open zhitao654321 opened 7 months ago

zhitao654321 commented 7 months ago

您好,我使用了您的模型对 COCO 数据集进行测试,发现我得到的评估指标与paper的指标有差距。我想了解一下,现在提供的demo和完整测试过程之间是否存在某些差异?

wjf5203 commented 5 months ago

Sorry for not replying in time. The weights provided previously for APP demo are the weights of the scale-up version, which are slightly lower than the joint-training version on COCO. The complete Model ZOO and inference scripts have been released, and the results in the paper can be fully reproduced, please try the new scripts~

HaisongDing commented 5 months ago

@zhitao654321 @wjf5203 Can you post the mAP of the scale-up version on COCO for reference?