From bf2f0bcfd1a7fbed462f65d44dd8589ab19ba715 Mon Sep 17 00:00:00 2001
From: natanielruiz <nataniel777@hotmail.com>
Date: 星期四, 26 十月 2017 03:19:35 +0800
Subject: [PATCH] Starting opensource
---
code/hopenet.py | 68 ++++++++++-----------------------
1 files changed, 21 insertions(+), 47 deletions(-)
diff --git a/code/hopenet.py b/code/hopenet.py
index c6bf0db..cd810e3 100644
--- a/code/hopenet.py
+++ b/code/hopenet.py
@@ -4,16 +4,6 @@
import math
import torch.nn.functional as F
-def ycbcr_to_rgb(input):
- # input is mini-batch N x 3 x H x W of an YCbCr image
- output = Variable(input.data.new(*input.size()))
- output[:, 0, :, :] = input[:, 0, :, :] + (input[:, 2, :, :] - 0.502) * 1.4
- output[:, 1, :, :] = input[:, 0, :, :] - (input[:, 1, :, :] - 0.502) * 0.343 - (input[:, 2, :, :] - 0.502) * 0.711
- output[:, 2, :, :] = input[:, 0, :, :] + (input[:, 1, :, :] - 0.502) * 1.765
- # output[output <= 0] = 0.
- # output[output > 1] = 1.
- return output
-
# CNN Model (2 conv layer)
class Simple_CNN(nn.Module):
def __init__(self):
@@ -235,20 +225,11 @@
roll = self.fc_roll(x)
return yaw, pitch, roll
-class Hopenet_SR(nn.Module):
+class Hopenet_new(nn.Module):
# This is just Hopenet with 3 output layers for yaw, pitch and roll.
- def __init__(self, block, layers, num_bins, upscale_factor):
+ def __init__(self, block, layers, num_bins):
self.inplanes = 64
- super(Hopenet, self).__init__()
- # Super resolution sub-network
- self.sr_relu = nn.ReLU()
- self.sr_conv1 = nn.Conv2d(1, 64, (5, 5), (1, 1), (2, 2))
- self.sr_conv2 = nn.Conv2d(64, 64, (3, 3), (1, 1), (1, 1))
- self.sr_conv3 = nn.Conv2d(64, 32, (3, 3), (1, 1), (1, 1))
- self.sr_conv4 = nn.Conv2d(32, upscale_factor ** 2, (3, 3), (1, 1), (1, 1))
- self.sr_pixel_shuffle = nn.PixelShuffle(upscale_factor)
-
- # Pose estimation sub-network
+ super(Hopenet_new, self).__init__()
self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,
bias=False)
self.bn1 = nn.BatchNorm2d(64)
@@ -264,9 +245,11 @@
self.fc_roll = nn.Linear(512 * block.expansion, num_bins)
self.softmax = nn.Softmax()
- self.idx_tensor = Variable(torch.FloatTensor(range(66))).cuda()
+ self.fc_finetune_new = nn.Linear(512 * block.expansion + 256 * block.expansion + 3, 3)
+ self.conv1x1 = nn.Conv2d(1024, 64, kernel_size = 1, stride = 1, bias=False)
+ self.maxpool_interm = nn.MaxPool2d(kernel_size=5, stride=3, padding=1)
- self.upscale_factor = upscale_factor
+ self.idx_tensor = Variable(torch.FloatTensor(range(66))).cuda()
for m in self.modules():
if isinstance(m, nn.Conv2d):
@@ -294,23 +277,7 @@
return nn.Sequential(*layers)
def forward(self, x):
- # Super-resolution sub-network
- y_channel = x[:,0,:,:]
-
- sr_y = self.sr_relu(self.sr_conv1(y_channel))
- sr_y = self.sr_relu(self.sr_conv2(sr_y))
- sr_y = self.sr_relu(self.sr_conv3(sr_y))
- sr_y = self.sr_pixel_shuffle(self.sr_conv4(sr_y))
-
- x[:,0,:,:] = sr_y
- x_rgb = ycbcr_to_rgb(x)
-
- out_img_cb = cb.resize(out_img_y.size, Image.BICUBIC)
- out_img_cr = cr.resize(out_img_y.size, Image.BICUBIC)
- out_img = Image.merge('YCbCr', [out_img_y, out_img_cb, out_img_cr]).convert('RGB')
-
- # Pose estimation sub-network
- x = self.conv1(sr_output)
+ x = self.conv1(x)
x = self.bn1(x)
x = self.relu(x)
x = self.maxpool(x)
@@ -318,6 +285,11 @@
x = self.layer1(x)
x = self.layer2(x)
x = self.layer3(x)
+ x_interm = self.conv1x1(x)
+ x_interm = self.relu(x_interm)
+ x_interm = self.maxpool_interm(x_interm)
+ x_interm = x_interm.view(x_interm.size(0), -1)
+
x = self.layer4(x)
x = self.avgpool(x)
@@ -327,16 +299,18 @@
pre_roll = self.fc_roll(x)
yaw = self.softmax(pre_yaw)
- yaw = Variable(torch.sum(yaw.data * self.idx_tensor.data, 1), requires_grad=True)
+ yaw = Variable(torch.sum(yaw.data * self.idx_tensor.data, 1), requires_grad=True) * 3 - 99
pitch = self.softmax(pre_pitch)
- pitch = Variable(torch.sum(pitch.data * self.idx_tensor.data, 1), requires_grad=True)
+ pitch = Variable(torch.sum(pitch.data * self.idx_tensor.data, 1), requires_grad=True) * 3 - 99
roll = self.softmax(pre_roll)
- roll = Variable(torch.sum(roll.data * self.idx_tensor.data, 1), requires_grad=True)
+ roll = Variable(torch.sum(roll.data * self.idx_tensor.data, 1), requires_grad=True) * 3 - 99
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 pre_yaw, pre_pitch, pre_roll, angles, sr_output
+ # angles predicts the residual
+ residuals = self.fc_finetune_new(torch.cat((preangles, x_interm, x), 1))
+ final_angles = preangles + residuals
+
+ return pre_yaw, pre_pitch, pre_roll, preangles, final_angles
--
Gitblit v1.8.0