search

fanq15 / FewX

FewX is an open-source toolbox on top of Detectron2 for data-limited instance-level recognition tasks.
https://github.com/fanq15/FewX
MIT License
346 stars 48 forks source link

I get AP 11.45. How to get AP 12? #39

Closed w39865008 closed 2 years ago

w39865008 commented 3 years ago

Hi，我在不同机器训练了两次，一次得到了11.45%的准确率，一次是11.40%，请问这个训练结果是不稳定吗，应该怎么得到 12% 的 AP 呢

fsod_train_log.txt fsod_finetune_test_log.txt fsod_finetune_train_log.txt

Zhang1Sheng commented 3 years ago

您好，我这边出现了一个参数报错的问题，是在fsod_roi_heads.py文件中，在_build_res5_block创建block时，因为first_stride这个参数报错。希望您能够帮忙解决一下问题，这是我的email：1434012475@qq.com ![Uploading PQMRE`9FTZHCP[4AE7}}WVO.png…]()

Zhang1Sheng commented 3 years ago

PQMRE`9FTZHCP 4AE7}}WVO

Zhang1Sheng commented 3 years ago

5JG6F2IO76BURVEO~251RB

Zhang1Sheng commented 3 years ago

Hi，我在不同机器训练了两次，一次得到了11.45%的准确率，一次是11.40%，请问这个训练结果是不稳定吗，应该怎么得到 12% 的 AP 呢

fsod_train_log.txt fsod_finetune_test_log.txt fsod_finetune_train_log.txt

(FNMNJ28(96TBAT}RV1 4K I know that the model generates 100 BBoxes for each class, but when I predicted that only dozens of BBoxes were generated for each class. Why not 100 and how to solve this problem?Could you help me out?

fanq15 commented 2 years ago

Hi, the testing performance is not stable. It is normal to get 11.4%. The reported performance in this repo is the best performance. You can refer to the performance in our paper for comparison.