From 9a02f63f4d5692399a95cb889e8f7629a165c28e Mon Sep 17 00:00:00 2001
From: natanielruiz <nataniel777@hotmail.com>
Date: 星期四, 21 九月 2017 05:56:20 +0800
Subject: [PATCH] next
---
code/hopenet.py | 113 +++++++++++++-------------------------------------------
1 files changed, 26 insertions(+), 87 deletions(-)
diff --git a/code/hopenet.py b/code/hopenet.py
index 1b94fa1..b02beec 100644
--- a/code/hopenet.py
+++ b/code/hopenet.py
@@ -1,8 +1,8 @@
import torch
import torch.nn as nn
-import torchvision.datasets as dsets
from torch.autograd import Variable
import math
+import torch.nn.functional as F
# CNN Model (2 conv layer)
class Simple_CNN(nn.Module):
@@ -41,7 +41,7 @@
class Hopenet(nn.Module):
# This is just Hopenet with 3 output layers for yaw, pitch and roll.
- def __init__(self, block, layers, num_bins):
+ def __init__(self, block, layers, num_bins, iter_ref):
self.inplanes = 64
super(Hopenet, self).__init__()
self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,
@@ -58,78 +58,12 @@
self.fc_pitch = nn.Linear(512 * block.expansion, num_bins)
self.fc_roll = nn.Linear(512 * block.expansion, num_bins)
- for m in self.modules():
- if isinstance(m, nn.Conv2d):
- n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
- m.weight.data.normal_(0, math.sqrt(2. / n))
- elif isinstance(m, nn.BatchNorm2d):
- m.weight.data.fill_(1)
- m.bias.data.zero_()
+ self.softmax = nn.Softmax()
+ self.fc_finetune = nn.Linear(512 * block.expansion + 3, 3)
- def _make_layer(self, block, planes, blocks, stride=1):
- downsample = None
- if stride != 1 or self.inplanes != planes * block.expansion:
- downsample = nn.Sequential(
- nn.Conv2d(self.inplanes, planes * block.expansion,
- kernel_size=1, stride=stride, bias=False),
- nn.BatchNorm2d(planes * block.expansion),
- )
+ self.idx_tensor = Variable(torch.FloatTensor(range(66))).cuda()
- layers = []
- layers.append(block(self.inplanes, planes, stride, downsample))
- self.inplanes = planes * block.expansion
- for i in range(1, blocks):
- layers.append(block(self.inplanes, planes))
-
- return nn.Sequential(*layers)
-
- def forward(self, x):
- x = self.conv1(x)
- x = self.bn1(x)
- x = self.relu(x)
- x = self.maxpool(x)
-
- x = self.layer1(x)
- x = self.layer2(x)
- x = self.layer3(x)
- x = self.layer4(x)
-
- x = self.avgpool(x)
- x = x.view(x.size(0), -1)
- yaw = self.fc_yaw(x)
- pitch = self.fc_pitch(x)
- roll = self.fc_roll(x)
-
- return yaw, pitch, roll
-
-class Hopenet_shape(nn.Module):
- # This is just Hopenet with 3 output layers for yaw, pitch and roll.
- def __init__(self, block, layers, num_bins, shape_bins):
- self.inplanes = 64
- super(Hopenet_shape, self).__init__()
- self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,
- bias=False)
- self.bn1 = nn.BatchNorm2d(64)
- self.relu = nn.ReLU(inplace=True)
- self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
- self.layer1 = self._make_layer(block, 64, layers[0])
- self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
- self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
- self.layer4 = self._make_layer(block, 512, layers[3], stride=2)
- self.avgpool = nn.AvgPool2d(7)
- self.fc_yaw = nn.Linear(512 * block.expansion, num_bins)
- self.fc_pitch = nn.Linear(512 * block.expansion, num_bins)
- self.fc_roll = nn.Linear(512 * block.expansion, num_bins)
- self.fc_shape_0 = nn.Linear(512 * block.expansion, shape_bins)
- self.fc_shape_1 = nn.Linear(512 * block.expansion, shape_bins)
- self.fc_shape_2 = nn.Linear(512 * block.expansion, shape_bins)
- self.fc_shape_3 = nn.Linear(512 * block.expansion, shape_bins)
- self.fc_shape_4 = nn.Linear(512 * block.expansion, shape_bins)
- self.fc_shape_5 = nn.Linear(512 * block.expansion, shape_bins)
- self.fc_shape_6 = nn.Linear(512 * block.expansion, shape_bins)
- self.fc_shape_7 = nn.Linear(512 * block.expansion, shape_bins)
- self.fc_shape_8 = nn.Linear(512 * block.expansion, shape_bins)
- self.fc_shape_9 = nn.Linear(512 * block.expansion, shape_bins)
+ self.iter_ref = iter_ref
for m in self.modules():
if isinstance(m, nn.Conv2d):
@@ -169,20 +103,25 @@
x = self.avgpool(x)
x = x.view(x.size(0), -1)
- yaw = self.fc_yaw(x)
- pitch = self.fc_pitch(x)
- roll = self.fc_roll(x)
+ pre_yaw = self.fc_yaw(x)
+ pre_pitch = self.fc_pitch(x)
+ pre_roll = self.fc_roll(x)
- shape = []
- shape.append(self.fc_shape_0(x))
- shape.append(self.fc_shape_1(x))
- shape.append(self.fc_shape_2(x))
- shape.append(self.fc_shape_3(x))
- shape.append(self.fc_shape_4(x))
- shape.append(self.fc_shape_5(x))
- shape.append(self.fc_shape_6(x))
- shape.append(self.fc_shape_7(x))
- shape.append(self.fc_shape_8(x))
- shape.append(self.fc_shape_9(x))
+ yaw = self.softmax(pre_yaw)
+ yaw = Variable(torch.sum(yaw.data * self.idx_tensor.data, 1), requires_grad=True)
+ pitch = self.softmax(pre_pitch)
+ pitch = Variable(torch.sum(pitch.data * self.idx_tensor.data, 1), requires_grad=True)
+ roll = self.softmax(pre_roll)
+ roll = Variable(torch.sum(roll.data * self.idx_tensor.data, 1), requires_grad=True)
+ yaw = yaw.view(yaw.size(0), 1)
+ pitch = pitch.view(pitch.size(0), 1)
+ roll = roll.view(roll.size(0), 1)
+ angles = []
+ preangles = torch.cat([yaw, pitch, roll], 1)
+ angles.append(preangles)
- return yaw, pitch, roll, shape
+ # angles predicts the residual
+ for idx in xrange(self.iter_ref):
+ angles.append(self.fc_finetune(torch.cat((preangles, x), 1)))
+
+ return pre_yaw, pre_pitch, pre_roll, angles
--
Gitblit v1.8.0