User Details
User Details
- User Since
- Aug 12 2016, 10:16 (460 w, 2 d)
- Availability
- Available
- Organization
- epfl.ch (university)
Feb 14 2017
Feb 14 2017
changes for final segmentation
Feb 8 2017
Feb 8 2017
Revert "missing reel"
missing reel
dezanet added a reverting change for R964:d8dc8e5f7d33: missing reel: R964:5cc87cd5ed4d: Revert "missing reel".
Feb 7 2017
Feb 7 2017
correct masking
wrong mask..
Feb 6 2017
Feb 6 2017
removed matplotlib
dezanet committed R964:ccbbedac5e80: in disparity cross link to last layer from input (authored by dezanet).
in disparity cross link to last layer from input
fixed test and unet
fixed testing
testing now segments everything
Feb 1 2017
Feb 1 2017
wrong part divided by 255
reduced kernel size
modified layout
bette output
removed regulariser on dilation/erosion
Jan 31 2017
Jan 31 2017
larger scale
dezanet committed R964:8a1a2f90b40f: dilation and erosion in different places (authored by dezanet).
dilation and erosion in different places
erosion layer
added nonlinearity
added linear combination
larger dilation
moved dilation
Jan 30 2017
Jan 30 2017
less regularisation
convolution and dilation
dezanet committed R964:ed11c9ede454: he_normal initialisation, RELU instead of ELU (authored by dezanet).
he_normal initialisation, RELU instead of ELU
removed augmentation
added mirror padding
valid convolution, l2 regularisation
Jan 19 2017
Jan 19 2017
add batch normalisation
add rotation again
back to original norm
no augmentation
different loss
new data normalization
add rotation augmentation
Jan 12 2017
Jan 12 2017
faster activation
Jan 10 2017
Jan 10 2017
back to net
Jan 9 2017
Jan 9 2017
more jumping
more filters, batchsize10
double filters
dilated test
Dec 22 2016
Dec 22 2016
removed reconstruction
Dec 5 2016
Dec 5 2016
fixed callback
flow net needs two outputs
more verbose output
dezanet committed R964:33bd35298d07: fixing problems with disparity regression (authored by dezanet).
fixing problems with disparity regression
Nov 30 2016
Nov 30 2016
wrong types
domain adaption with new model
Nov 29 2016
Nov 29 2016
use float images
moar filters
moved normalise to data.py
normalization in test
double width
more conv per level
make data great again
augmentation failed
augmentation
increased learning rate
return correct sized image
moving the data to -100,100
Nov 28 2016
Nov 28 2016
no normalisation
no ReLU in first layer
corrected normalisation
different normalisation
reduced batch size
normalisation bug
try normalisation
slower learning
no normalisation
small cleanup
remove generator
included batch size
callback images generation
no augmentation
increased number of samples
reduced amount of augmentation
proper augmentation
Nov 25 2016
Nov 25 2016
back to unit
Nov 22 2016
Nov 22 2016
increased recursion limit
naming correction
more convolutions
batchsize=1
no identity blocks