-
Notifications
You must be signed in to change notification settings - Fork 4
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Unfair experimental settings. (DeepLab v3+ vs. DeepLab v2) #3
Comments
Yes please update the results. |
For the warmup you have used AdaptNet which uses DeeplabV2 as the Segmentation model, was that also changed to DeeplabV3+? If yes can you please report the initial result after warmup? |
Sorry for the delay. @zhijiew @sharat29ag Our training settings is based on the CAG_UDA (Nips 2019, https://github.com/RogerZhangzz/CAG_UDA). However, the author has written the DeeplabV3+ to DeeplabV2. The only thing I can do is to report my result and framework honestly. |
The reported warmup model is a DeeplabV3+ model. For a fair comparison, I reused the warmup model from the CAG_UDA (Nips 2019, https://github.com/RogerZhangzz/CAG_UDA), just here The origin report is here: I hope my reply can dispel your doubts @sharat29ag . |
Thank you for the response will look to it. |
Is there any mistake, I use the model of DeepLabV3+ from https://github.com/RogerZhangzz/CAG_UDA to train cityscape, and got the following result:
I using amp to save memory so that I can train the model with single 2080ti. |
Hi, @munanning please share the initial weights for V2. Also the selection list contains 297 images, please share list of 150 images for fair comparison. |
@sharat29ag |
Thanks. |
@munanning does the warmup weights at sgate1 for GTA->City and Synthia->City were same? |
In your experiments, your method used DeepLab v3+ as the backbone to compare with other methods that use DeepLab v2, which is totally unfair. Can you report the results based on DeepLab v2?
The text was updated successfully, but these errors were encountered: