spp

.vscode/tags

@@ -0,0 +1,23 @@
+!_TAG_FILE_FORMAT	2	/extended format; --format=1 will not append ;" to lines/
+!_TAG_FILE_SORTED	1	/0=unsorted, 1=sorted, 2=foldcase/
+!_TAG_PROGRAM_AUTHOR	Darren Hiebert	/[email protected]/
+!_TAG_PROGRAM_NAME	Exuberant Ctags	//
+!_TAG_PROGRAM_URL	http://ctags.sourceforge.net	/official site/
+!_TAG_PROGRAM_VERSION	5.8	//
+F	../cnn_with_spp.py	/^import torch.nn.functional as F$/;"	kind:namespace	line:7
+SPP_NET	../cnn_with_spp.py	/^class SPP_NET(nn.Module):$/;"	kind:class	line:9
+Variable	../cnn_with_spp.py	/^from torch.autograd import Variable$/;"	kind:namespace	line:5
+__init__	../cnn_with_spp.py	/^    def __init__(self, opt, input_nc, ndf=64,  gpu_ids=[]):$/;"	kind:member	line:13
+cnn_with_spp.py	../cnn_with_spp.py	1;"	kind:file	line:1
+forward	../cnn_with_spp.py	/^    def forward(self,x):$/;"	kind:member	line:33
+functools	../cnn_with_spp.py	/^import functools$/;"	kind:namespace	line:4
+init	../cnn_with_spp.py	/^from torch.nn import init$/;"	kind:namespace	line:3
+nn	../cnn_with_spp.py	/^import torch.nn as nn$/;"	kind:namespace	line:2
+nn	../cnn_with_spp.py	/^import torch.nn.functional as F$/;"	kind:namespace	line:7
+np	../cnn_with_spp.py	/^import numpy as np$/;"	kind:namespace	line:6
+spatial_pyramid_pool	../cnn_with_spp.py	/^from spp_layer import spatial_pyramid_pool$/;"	kind:namespace	line:8
+spatial_pyramid_pool	../spp_layer.py	/^def spatial_pyramid_pool(self,previous_conv, num_sample, previous_conv_size, out_pool_size):$/;"	kind:function	line:1
+spp_layer.py	../spp_layer.py	1;"	kind:file	line:1
+torch	../cnn_with_spp.py	/^import torch$/;"	kind:namespace	line:1
+torch	../cnn_with_spp.py	/^import torch.nn as nn$/;"	kind:namespace	line:2
+torch	../cnn_with_spp.py	/^import torch.nn.functional as F$/;"	kind:namespace	line:7

cnn_with_spp.py

@@ -26,7 +26,7 @@ def __init__(self, opt, input_nc, ndf=64,  gpu_ids=[]):
         self.conv4 = nn.Conv2d(ndf * 4, ndf * 8, 4, 1, 1, bias=False)
         self.BN3 = nn.BatchNorm2d(ndf * 8)
 
-        self.conv5 = nn.Conv2d(ndf * 8, 1, 4, 1, 0, bias=False)
+        self.conv5 = nn.Conv2d(ndf * 8, 64, 4, 1, 0, bias=False)
         self.fc1 = nn.Linear(10752,4096)
         self.fc2 = nn.Linear(4096,1000)
 

spp_layer.py

@@ -1,3 +1,4 @@
+import math
 def spatial_pyramid_pool(self,previous_conv, num_sample, previous_conv_size, out_pool_size):
     '''
     previous_conv: a tensor vector of previous convolution layer
@@ -10,11 +11,11 @@ def spatial_pyramid_pool(self,previous_conv, num_sample, previous_conv_size, out
     # print(previous_conv.size())
     for i in range(len(out_pool_size)):
         # print(previous_conv_size)
-        h_strd = previous_conv_size[0] / out_pool_size[i]
-        w_strd = previous_conv_size[1] / out_pool_size[i]
-        h_wid = previous_conv_size[0] - h_strd * out_pool_size[i] + 1
-        w_wid = previous_conv_size[1] - w_strd * out_pool_size[i] + 1
-        maxpool = nn.MaxPool2d((h_wid,w_wid),stride=(h_strd,w_strd))
+        h_wid = int(math.ceil(previous_conv_size[0] / out_pool_size[i]))
+        w_wid = int(math.ceil(previous_conv_size[1] / out_pool_size[i]))
+        h_pad = (h_wid*out_pool_size[i] - previous_conv_size[0] + 1)/2
+        w_pad = (w_wid*out_pool_size[i] - previous_conv_size[1] + 1)/2
+        maxpool = nn.MaxPool2d((h_wid, w_wid), stride=(h_wid, w_wid), padding=(h_pad, w_pad))
         x = maxpool(previous_conv)
         if(i == 0):
             spp = x.view(num_sample,-1)