search

johnnylu305 / Simple-does-it-weakly-supervised-instance-and-semantic-segmentation

Weakly Supervised Segmentation by Tensorflow. Implements semantic segmentation in Simple Does It: Weakly Supervised Instance and Semantic Segmentation, by Khoreva et al. (CVPR 2017).
Apache License 2.0
55 stars 12 forks source link

效果怎么和原论文差这么多？ #2

Closed kilaOoO closed 5 years ago

johnnylu305 commented 5 years ago

In my opinion, I don't use M∩G+ to generate the labels for the network, so the result isn't great enough.

kilaOoO commented 5 years ago

我也觉得是这个原因，但主要差的有点多，因为在我看来差距可能不会这么大，所以才想咨询一下。

At 2019-07-18 21:10:29, "CHENG-YOU LU" notifications@github.com wrote:

In my opinion, I don't use M∩G+ to generate the labels for the network, so the result isn't great enough.

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

johnnylu305 commented 5 years ago

Hello, @kilaOoO. Yes, M∩G+ is the main reason why the performance is not good. However, there are several minor reasons. For example, they use early stopping and large batch sizes. Therefore, I can't follow all the parameters they gave because of the different structures.