vikolss / DACS

Code from the paper "DACS: Domain Adaptation via Cross-domain Mixed Sampling"
MIT License
123 stars 19 forks source link

can not connect to download the pretrained model #15

Closed ywher closed 2 years ago

ywher commented 2 years ago

hi, everyone. I have a problem of connecting to http://vllab1.ucmerced.edu/~whung/adv-semi-seg/resnet101COCO-41f33a49.pth to download the pretrained model. Can you give me a help by uploading to google drive?

vikolss commented 2 years ago

Thank you for bringing this up! I am not the owner of that link, and I am also not able to reach it. Did you manage to reach it or find the model elsewhere?

ywher commented 2 years ago

Thank you for bringing this up! I am not the owner of that link, and I am also not able to reach it. Did you manage to reach it or find the model elsewhere?

Sorry, I still can not connect to that URL. But I search the model name 'resnet101COCO-41f33a49.pth' and find the public code 'https://github.com/hb-stone/FC-SOD' also contains the pretrained model of ResNet101 on COCO which is also named resnet101COCO-41f33a49.pth. The link of the model is https://drive.google.com/file/d/1gLiF0yKByduIyD7MEya-4ZI_yp5613Ot/view?usp=sharing. I think they maybe the same, but not so sure.

vikolss commented 2 years ago

Ok, nice find. I agree that it seems that they might be the same, they at least have the same architecture. I am unfortunately not able to perform any tests at the moment to see if the performance is the same with this model as with the original one. If you do, please let me know how it went :)

ywher commented 2 years ago

Ok, nice find. I agree that it seems that they might be the same, they at least have the same architecture. I am unfortunately not able to perform any tests at the moment to see if the performance is the same with this model as with the original one. If you do, please let me know how it went :)

I have successfully reproduce the result using this pretrained model. After training 25k iters: class 0 road IU 91.49 class 1 sidewalk IU 36.92 class 2 building IU 87.71 class 3 wall IU 27.02 class 4 fence IU 38.51 class 5 pole IU 39.64 class 6 traffic_light IU 43.36 class 7 traffic_sign IU 50.61 class 8 vegetation IU 87.78 class 9 terrain IU 40.95 class 10 sky IU 89.38 class 11 person IU 65.97 class 12 rider IU 29.13 class 13 car IU 90.30 class 14 truck IU 58.66 class 15 bus IU 54.12 class 16 train IU 0.00 class 17 motorcycle IU 37.28 class 18 bicycle IU 33.92 meanIOU: 0.527782650365583

vikolss commented 2 years ago

That's great! Thank you very much!

mkbkxdd commented 2 years ago

Ok, nice find. I agree that it seems that they might be the same, they at least have the same architecture. I am unfortunately not able to perform any tests at the moment to see if the performance is the same with this model as with the original one. If you do, please let me know how it went :)

I have successfully reproduce the result using this pretrained model. After training 25k iters: class 0 road IU 91.49 class 1 sidewalk IU 36.92 class 2 building IU 87.71 class 3 wall IU 27.02 class 4 fence IU 38.51 class 5 pole IU 39.64 class 6 traffic_light IU 43.36 class 7 traffic_sign IU 50.61 class 8 vegetation IU 87.78 class 9 terrain IU 40.95 class 10 sky IU 89.38 class 11 person IU 65.97 class 12 rider IU 29.13 class 13 car IU 90.30 class 14 truck IU 58.66 class 15 bus IU 54.12 class 16 train IU 0.00 class 17 motorcycle IU 37.28 class 18 bicycle IU 33.92 meanIOU: 0.527782650365583

大兄弟 ,我拿作者的模型(miou53)跑出来的结果却是这样的,但我不知道哪里出问题了,你能帮一下我吗,谢谢了 class 0 road IU 67.69 class 1 sidewalk IU 9.53 class 2 building IU 42.74 class 3 wall IU 3.70 class 4 fence IU 5.86 class 5 pole IU 1.79 class 6 traffic_light IU 0.83 class 7 traffic_sign IU 2.28 class 8 vegetation IU 37.00 class 9 terrain IU 10.66 class 10 sky IU 11.93 class 11 person IU 5.54 class 12 rider IU 0.38 class 13 car IU 26.87 class 14 truck IU 2.51 class 15 bus IU 5.62 class 16 train IU 0.00 class 17 motorcycle IU 3.78 class 18 bicycle IU 11.55 meanIOU: 0.13172500007111895