Submitted by Caner Hazirbas.

Submission data

Full nameFuseNet-SparseFusion5
DescriptionWe train the FuseNet-SF5 on the ScanNet v2 dataset. During training, we resize the images and depth maps to 240x320 and upsample the results with the nearest-neighbor interpolation to the full resolution during test.
Publication titleFuseNet: Incorporating Depth into Semantic Segmentation via Fusion-based CNN Architecture
Publication authorsCaner Hazirbas, Lingni Ma, Csaba Domokos, Daniel Cremers
Publication venueACCV 2016
Publication URLhttps://link.springer.com/chapter/10.1007/978-3-319-54181-5_14
Input Data TypesUses Color,Uses Geometry        Uses 2D
Programming language(s)Python with CUDA
HardwareIntelĀ® Xeon(R) CPU E5-2623 v3, GeForce TITAN X (Pascal)
Websitehttps://hazirbas.com/projects/fusenet/
Source code or download URLhttps://github.com/MehmetAygun/fusenet-pytorch
Submission creation date3 Dec, 2018
Last edited3 Dec, 2018

2D semantic label results

Infoavg ioubathtubbedbookshelfcabinetchaircountercurtaindeskdoorfloorotherfurniturepicturerefrigeratorshower curtainsinksofatabletoiletwallwindow
permissive0.5210.5910.6820.2200.4880.2790.3440.6100.4610.4750.9100.2930.4470.5120.3970.6180.5670.4520.7340.7820.566