search

LikeLy-Journey / SegmenTron

Support PointRend, Fast_SCNN, HRNet, Deeplabv3_plus(xception, resnet, mobilenet), ContextNet, FPENet, DABNet, EdaNet, ENet, Espnetv2, RefineNet, UNet, DANet, HRNet, DFANet, HardNet, LedNet, OCNet, EncNet, DuNet, CGNet, CCNet, BiSeNet, PSPNet, ICNet, FCN, deeplab)
Apache License 2.0
705 stars 162 forks source link

Xception pretrained #23

Open haibochina opened 4 years ago

haibochina commented 4 years ago

Thank you for your great job! Is there any pretrained model for DFANet?

LikeLy-Journey commented 4 years ago

i have not trained this model

haibochina commented 4 years ago

i have not trained this model

I am training now!

linbaiwpi commented 4 years ago

i have not trained this model

I am training now!

How is your training right now? Using the pre-trained Xception or training from scratch?

haibochina commented 4 years ago

It is still training now！ The result without pretrained model is 59 miou 

发自我的iPhone

------------------ Original ------------------ From: Lin Bai <notifications@github.com> Date: Sat,Mar 14,2020 9:57 AM To: LikeLy-Journey/SegmenTron <SegmenTron@noreply.github.com> Cc: haibo <1974273017@qq.com>, Author <author@noreply.github.com> Subject: Re: [LikeLy-Journey/SegmenTron] Xception pretrained (#23)

i have not trained this model

I am training now!

How is your training right now? Using the pre-trained Xception or training from scratch?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

haliphinx commented 4 years ago

@haibochina Hey, are you training the DFANet on the cityscape dataset? What's your final accuracy and what's the number of epoch in total to achieve that? I trained it from scratch, but it stuck into around 30 miou from epoch 100. Thanks!!

haibochina commented 4 years ago

@haliphinx I just trained XceptionA the way that shufflenet use. and just trained 30000 iterarion, then I trainied DFANet on cityscape, the mIOU is 68.9. https://github.com/megvii-model/ShuffleNet-Series

haliphinx commented 4 years ago

@haibochina Thanks for your reply. The accuracy sounds great! I'm wondering if it's possible to share your pretrained XceptionA model please.