search

facebookresearch / unbiased-teacher

PyTorch code for ICLR 2021 paper Unbiased Teacher for Semi-Supervised Object Detection
https://arxiv.org/abs/2102.09480
MIT License
409 stars 84 forks source link

COCOEval on Pascal dataset experiments #40

Closed JongMokKim closed 2 years ago

JongMokKim commented 2 years ago

Hi,

I saw your README.MD that there is an higher AP issue when using voc_evaluator than coco_evaluator.

I just want to reproduce your results of paper, so I want to train voc data and evaluate network on coco_evaluator

I used your config script that includes EVALUATOR : COCOEVAL, but it seems voc evaluator ran, not coco evaluator.

Could you share your way?

ycliu93 commented 2 years ago

Sorry, the EVALUATOR in config file is deprecated. If you would like to change the evaluator to the original evaluator.

Please change this line
https://github.com/facebookresearch/unbiased-teacher/blob/6977c6f77c812fae4064dc1b3865658c2ed247b1/ubteacher/engine/trainer.py#L166

to 

return COCOEvaluator(dataset_name, output_dir=output_folder)

In this way, you are able to use the COCO evaluator.