search

MehmetAygun / fusenet-pytorch

Other
80 stars 17 forks source link

worser test result on nyu dataset #10

Closed hurricane2018 closed 5 years ago

hurricane2018 commented 5 years ago

First, I use the default code to train the model on nyu dataset. Second, I test the nyu dataset using the default code. However, I get a very bad result as shown in the following figure. Could you give me some suggestion?

image

Best,

hurricane2018 commented 5 years ago

@MehmetAygun @hazirbas Can you help me ?

MehmetAygun commented 5 years ago

Hey @hurricane2018, it seems you used the last saved checkpoint for testing, which can be not a very good model. Can you look your log, and find which checkpoint has lower loss, and test using that model.

Alice-kenan commented 5 years ago

Hi, Can you provide more information for training the model on caffe?about how to get the rgb-d lmdb data. Looking forward to your reply!

MehmetAygun commented 5 years ago

@Alice-kenan for caffe implementation please refer to : https://github.com/tum-vision/fusenet

Alice-kenan commented 5 years ago

@MehmetAygun OK, thanks for your reply!