search

xingyizhou / ExtremeNet

Bottom-up Object Detection by Grouping Extreme and Center Points
BSD 3-Clause "New" or "Revised" License
1.03k stars 172 forks source link

How to train without the pre-trained cornernet model #8

Closed XiaoR666 closed 5 years ago

XiaoR666 commented 5 years ago

thank u for your code very much，i don not want to fine-tun from the cornernet model and how to train the model from scratch

xingyizhou commented 5 years ago

Hi, Thanks for your interests. You can delete this line https://github.com/xingyizhou/ExtremeNet/blob/master/config/ExtremeNet.json#L26 to train from scratch. To fully reproduce our results, you will also need to increase the training iterations https://github.com/xingyizhou/ExtremeNet/blob/master/config/ExtremeNet.json#L18 . We have tried "max_iter": 500000, "stepsize": 450000, it works slightly better than fine-tuning CornerNet. Our model is given at https://drive.google.com/file/d/1omiOUjWCrFbTJREypuZaODu0bOlF_7Fg/view?usp=sharing .

XiaoR666 commented 5 years ago

Thanks very much，it help me a lot.