zanilzanzan / FuseNet_PyTorch

Joint scene classification and semantic segmentation with FuseNet
GNU General Public License v3.0
108 stars 34 forks source link

sceneType issue #4

Closed Aleck16 closed 6 years ago

Aleck16 commented 6 years ago

Hello, The original dataset nyu_depth_v2_labeled.mat contains 26 scene categories. Your nyu_class_10_db.h5 dataset has only 10 scene categories. I want to know how you extracted these 10 scene categories? Also, the Datasets SUNRGBD dataset download link you provided is invalid, I don't know where to get it.

hazirbas commented 6 years ago

hey, I uploaded the mapping file under utils folder. SUNRGB-D will be ready later.

hazirbas commented 6 years ago

Dataset is added.

shangweihung commented 6 years ago

Hi, does your uploaded SUNRGB-D files also adopt class mapping? How many classes after the mapping?

I run 'Ok' the code with the NYU data you provided, but got wrong message as below no matter how I change the variable 'num_classes' in the FuseNetClass.py

image

zanilzanzan commented 6 years ago

Hey @shangweihung, the uploaded version of SUN RGB-D unfortunately doesn’t include scene class labels. I continued my experiments with only NYU v2 after one point, so didn’t feel the need to use class mapping for SUN. I’ll try to prepare it by the weekend. Till then you can run the vanilla version of FuseNet with SUN, by just changing the imported model and solver.