search

chengchunhsu / EveryPixelMatters

Implementation of ECCV 2020 paper "Every Pixel Matters: Center-aware Feature Alignment for Domain Adaptive Object Detector"
Other
164 stars 21 forks source link

can you provide the training log of cityscapes to foggy_cityscapes? #5

Open 0Godness opened 3 years ago

0Godness commented 3 years ago

hi,
Thanks for your open code! my results have some difference from this paper. and I think that my configs may be wrong, so I want to look at your training log of C to F. Thank you again!

SkeletonOne commented 3 years ago

I am also looking forward to a log file (or maybe two, because I think training GA and CA are separated) about the whole training procedure. The total iters of the training seem to be not clear. How many iters should I train the GA and how many iters on the GA+CA?

0Godness commented 3 years ago

I am also looking forward to a log file (or maybe two, because I think training GA and CA are separated) about the whole training procedure. The total iters of the training seem to be not clear. How many iters should I train the GA and how many iters on the GA+CA?

Different datasets need different training iters. Take KITTI as an example, it needs 3k iters for GA and 5K iters for GA+CA with 2k pretrain iters. I suggest that you should try a few iterations around this number.

SkeletonOne commented 3 years ago

@0Godness Thanks for your information. Actually, I did not change the GA+CA (for example, in Kitti, 10k iters for ga+ca with 5k pre-trained ga) in the configs. I am just wondering if the iters are sufficient for achieving the best performance or damage the performance in turn.

0Godness commented 3 years ago

@SkeletonOne In my experiments, the iters are a little sensitive for the performance which may have 1AP error because of the different iters.

mochaojie commented 3 years ago

hi, Thanks for your open code! my results have some difference from this paper. and I think that my configs may be wrong, so I want to look at your training log of C to F. Thank you again!

你好，请问foggy cityscapes的数据集怎么下啊，我进官网看，看它的介绍数据集好像都是给语义分割准备的或者是自动驾驶，我不知道下哪个，你能不能具体告诉我下哪个

vzy2d commented 3 years ago

hi, Thanks for your open code! my results have some difference from this paper. and I think that my configs may be wrong, so I want to look at your training log of C to F. Thank you again!

你好，请问foggy cityscapes的数据集怎么下啊，我进官网看，看它的介绍数据集好像都是给语义分割准备的或者是自动驾驶，我不知道下哪个，你能不能具体告诉我下哪个

从cityscapes官网下

sunshehai commented 1 year ago

Do you meet the error: In the last step of installing：python setup.py build develop , I encountered problem：Couldn t find a setup script in /tmp/easy_install-2xsxsqeq/scikit_image-0.20.0.tar.gz. Thank you @0Godness @vzy2d