ViT-L (66.0 AP) pre-training & config file

Sense-X / Co-DETR

[ICCV 2023] DETRs with Collaborative Hybrid Assignments Training

MIT License

969 stars 105 forks source link

ViT-L (66.0 AP) pre-training & config file #37

Open RicoJYang opened 1 year ago

RicoJYang commented 1 year ago

请问ViT-L (66.0 AP)的模型的backbone是用的eva02的det任务给出的eva02_L_pt_m38m_p14to16 | 304M | Merged-38M | 56这个预训练模型嘛，如果不是是否可以说明一下具体使用的模型。论文中提到的Co-DINO-Deformable-DETR各部分是如何组成的呢，感谢！

TempleX98 commented 1 year ago

pretrain: eva02_L_pt_m38m_medft_in21k_ft_in1k_p14 The model is Co-DINO and more details about this large model are presented in the paper appendix.

RicoJYang commented 1 year ago

pretrain: eva02_L_pt_m38m_medft_in21k_ft_in1k_p14 The model is Co-DINO and more details about this large model are presented in the paper appendix. 感谢您的回答，Co-DINO 这个配置文件似乎是基于Swin Transformer的64.1性能的模型，如果要复现ViT-L (66.0 AP)的话，是否是需要修改其中的backbone等部分，如果方便您可以分享对应的config file 和 backbone file嘛，再次表示感谢！