Skip to content

small change for PyTorch v1.5, Python 3.7 #31

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
leekunhee wants to merge 1 commit into
base: master
Choose a base branch
from
Open

small change for PyTorch v1.5, Python 3.7 #31

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
32 changes: 13 additions & 19 deletions depthNet_model.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,20 +6,14 @@
import torch.nn.functional as F
import torch.nn.init as init
from torch.autograd import Variable
from torch import Tensor
import cv2
import math
import numpy as np
import time
from numpy.linalg import inv

def down_conv_layer(input_channels, output_channels, kernel_size):
return nn.Sequential(
nn.Conv2d(
input_channels,
output_channels,
kernel_size,
padding=(kernel_size - 1) / 2,
padding=(kernel_size - 1) // 2,
stride=1,
bias=False),
nn.BatchNorm2d(output_channels),
Expand All @@ -28,7 +22,7 @@ def down_conv_layer(input_channels, output_channels, kernel_size):
output_channels,
output_channels,
kernel_size,
padding=(kernel_size - 1) / 2,
padding=(kernel_size - 1) // 2,
stride=2,
bias=False),
nn.BatchNorm2d(output_channels),
Expand All @@ -40,7 +34,7 @@ def conv_layer(input_channels, output_channels, kernel_size):
input_channels,
output_channels,
kernel_size,
padding=(kernel_size - 1) / 2,
padding=(kernel_size - 1) // 2,
bias=False),
nn.BatchNorm2d(output_channels),
nn.ReLU())
Expand All @@ -54,12 +48,12 @@ def refine_layer(input_channels):

def up_conv_layer(input_channels, output_channels, kernel_size):
return nn.Sequential(
nn.Upsample(scale_factor=2, mode='bilinear'),
nn.Upsample(scale_factor=2, mode='bilinear', align_corners=True),
nn.Conv2d(
input_channels,
output_channels,
kernel_size,
padding=(kernel_size - 1) / 2,
padding=(kernel_size - 1) // 2,
bias=False),
nn.BatchNorm2d(output_channels),
nn.ReLU())
Expand Down Expand Up @@ -114,15 +108,15 @@ def __init__(self):
total_num = 0
for m in self.modules():
if isinstance(m, nn.Conv2d):
init.kaiming_normal(m.weight, mode='fan_out')
init.kaiming_normal_(m.weight, mode='fan_out')
total_num += get_trainable_number(m.weight)
if m.bias is not None:
init.constant(m.bias, 0)
init.constant_(m.bias, 0)
total_num += get_trainable_number(m.bias)
elif isinstance(m, nn.BatchNorm2d):
init.constant(m.weight, 1)
init.constant_(m.weight, 1)
total_num += get_trainable_number(m.weight)
init.constant(m.bias, 0)
init.constant_(m.bias, 0)
total_num += get_trainable_number(m.bias)
elif isinstance(m, nn.Linear):
init.normal(m.weight, std=1e-3)
Expand Down Expand Up @@ -159,7 +153,7 @@ def getVolume(self, left_image, right_image, KRKiUV_T, KT_T):

warp_uv = Variable(warp_uv.permute(
0, 3, 2, 1)) #shape = batch x height x width x 2
warped = F.grid_sample(right_image, warp_uv)
warped = F.grid_sample(right_image, warp_uv, align_corners=True)

costvolume[:, depth_i, :, :] = torch.sum(
torch.abs(warped - left_image), dim=1)
Expand All @@ -182,17 +176,17 @@ def forward(self, left_image, right_image, KRKiUV_T, KT_T):
upconv4 = self.upconv4(iconv5)
iconv4 = self.iconv4(torch.cat((upconv4, conv3), 1))
disp4 = 2.0 * self.disp4(iconv4)
udisp4 = F.upsample(disp4, scale_factor=2)
udisp4 = F.interpolate(disp4, scale_factor=2)

upconv3 = self.upconv3(iconv4)
iconv3 = self.iconv3(torch.cat((upconv3, conv2, udisp4), 1))
disp3 = 2.0 * self.disp3(iconv3)
udisp3 = F.upsample(disp3, scale_factor=2)
udisp3 = F.interpolate(disp3, scale_factor=2)

upconv2 = self.upconv2(iconv3)
iconv2 = self.iconv2(torch.cat((upconv2, conv1, udisp3), 1))
disp2 = 2.0 * self.disp2(iconv2)
udisp2 = F.upsample(disp2, scale_factor=2)
udisp2 = F.interpolate(disp2, scale_factor=2)

upconv1 = self.upconv1(iconv2)
iconv1 = self.iconv1(torch.cat((upconv1, udisp2), 1))
Expand Down
15 changes: 8 additions & 7 deletions example.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,7 +13,7 @@
from visualize import *

with open('sample_data.pkl', 'rb') as fp:
sample_datas = pickle.load(fp)
sample_datas = pickle.load(fp, encoding='latin1')

# model
depthnet = depthNet()
Expand All @@ -35,14 +35,15 @@
for this_sample in sample_datas:

# get data
depth_image_cuda = Tensor(this_sample['depth_image']).cuda()
depth_image_cuda = Variable(depth_image_cuda, volatile=True)
with torch.no_grad():
depth_image_cuda = Tensor(this_sample['depth_image']).cuda()
depth_image_cuda = Variable(depth_image_cuda)

left_image_cuda = Tensor(this_sample['left_image']).cuda()
left_image_cuda = Variable(left_image_cuda, volatile=True)
left_image_cuda = Tensor(this_sample['left_image']).cuda()
left_image_cuda = Variable(left_image_cuda)

right_image_cuda = Tensor(this_sample['right_image']).cuda()
right_image_cuda = Variable(right_image_cuda, volatile=True)
right_image_cuda = Tensor(this_sample['right_image']).cuda()
right_image_cuda = Variable(right_image_cuda)

left_in_right_T = this_sample['left2right'][0:3, 3]
left_in_right_R = this_sample['left2right'][0:3, 0:3]
Expand Down