Huan Liu

Why GeometryEncoder instead GlobalGenerator

3

Hello, nice job! I have a question, why don't use the encoder(including downsample and resnet-blocks) part of ``GlobalGenerator`` with ngf=32 as Geometry Encoder directly, but train a ``GeometryEncoder`` with ngf=64....

Thank you for sharing your work. Recently, I encountered some problems in reproducing your PETR results. Can you share your training logs with the config of `petr_r50_16x2_100e_coco.py` and the corresponding...

Hi~ Thanks for this wonderful project. Currently, I try to perform interactive annotation for automatically polishing the description of each bbox region via a LLM, which looks like ``` ```...