search

megvii-model / YOLOF

MIT License
809 stars 115 forks source link

request for yolof-dc5 model #28

Open emjay73 opened 3 years ago

emjay73 commented 3 years ago

Can I get a model config and pre-trained weights of the model denoted in this table? image

chensnathan commented 3 years ago

Hi, we do not have a plan to provide the 3-rd stage model now. Maybe next month. I'll keep this issue open.

You can train the third stage on your own with the provided 2-nd stage pre-trained model. Please follow the steps in this repo.

emjay73 commented 3 years ago

You mean the only difference between the YOLOF_CSP_D_53_DC5_9x_stage2_3x and YOLOF-DC5 in the above table is whether the SWA phase is applied or not? Then where does the fps difference come from?

chensnathan commented 3 years ago

You can check the caption of the Table. When comparing with YOLOv4, we first fuse conv and bn, then keep cudnn_benchmark as True, and use half-precision for inference.

emjay73 commented 3 years ago

Which means, 

SOLVER.AMP.ENABLED : true 
CUDNN_BENCHMARK : true

in the config file?