Fix #125 and #154

Author: adam9500370

ptsemseg/loss.py

@@ -11,10 +11,10 @@ def cross_entropy2d(input, target, weight=None, size_average=True):
     # Handle inconsistent size between input and target
     if h > ht and w > wt:  # upsample labels
         target = target.unsqueeze(1)
-        target = F.upsample(target, size=(h, w), mode="nearest")
+        target = F.interpolate(target.float(), size=(h, w), mode="nearest").long()
         target = target.squeeze(1)
     elif h < ht and w < wt:  # upsample images
-        input = F.upsample(input, size=(ht, wt), mode="bilinear")
+        input = F.interpolate(input, size=(ht, wt), mode="bilinear", align_corners=True)
     elif h != ht and w != wt:
         raise Exception("Only support upsampling")
 
@@ -72,9 +72,7 @@ def multi_scale_cross_entropy2d(
     if scale_weight == None:  # scale_weight: torch tensor type
         n_inp = len(input)
         scale = 0.4
-        scale_weight = torch.pow(scale * torch.ones(n_inp), torch.arange(n_inp))
-        if input.is_cuda:
-            scale_weight = scale_weight.cuda()
+        scale_weight = torch.pow(scale * torch.ones(n_inp), torch.arange(n_inp).float()).to('cuda' if input.is_cuda else 'cpu')
 
     loss = 0.0
     for i, inp in enumerate(input):

ptsemseg/loss/loss.py

@@ -11,10 +11,10 @@ def cross_entropy2d(input, target, weight=None, size_average=True):
     # Handle inconsistent size between input and target
     if h > ht and w > wt:  # upsample labels
         target = target.unsequeeze(1)
-        target = F.upsample(target, size=(h, w), mode="nearest")
+        target = F.interpolate(target.float(), size=(h, w), mode="nearest").long()
         target = target.sequeeze(1)
     elif h < ht and w < wt:  # upsample images
-        input = F.upsample(input, size=(ht, wt), mode="bilinear")
+        input = F.interpolate(input, size=(ht, wt), mode="bilinear", align_corners=True)
     elif h != ht and w != wt:
         raise Exception("Only support upsampling")
 
@@ -33,7 +33,7 @@ def multi_scale_cross_entropy2d(
     if scale_weight == None:  # scale_weight: torch tensor type
         n_inp = len(input)
         scale = 0.4
-        scale_weight = torch.pow(scale * torch.ones(n_inp), torch.arange(n_inp))
+        scale_weight = torch.pow(scale * torch.ones(n_inp), torch.arange(n_inp).float()).to('cuda' if input.is_cuda else 'cpu')
 
     loss = 0.0
     for i, inp in enumerate(input):

ptsemseg/models/icnet.py

@@ -179,7 +179,7 @@ def forward(self, x):
         h, w = x.shape[2:]
 
         # H, W -> H/2, W/2
-        x_sub2 = interp(x, output_size=get_interp_size(x, s_factor=2))
+        x_sub2 = F.interpolate(x, size=get_interp_size(x, s_factor=2), mode='bilinear', align_corners=True)
 
         # H/2, W/2 -> H/4, W/4
         x_sub2 = self.convbnrelu1_1(x_sub2)
@@ -193,7 +193,7 @@ def forward(self, x):
         x_sub2 = self.res_block2(x_sub2)
         x_sub2 = self.res_block3_conv(x_sub2)
         # H/16, W/16 -> H/32, W/32
-        x_sub4 = interp(x_sub2, output_size=get_interp_size(x_sub2, s_factor=2))
+        x_sub4 = F.interpolate(x_sub2, size=get_interp_size(x_sub2, s_factor=2), mode='bilinear', align_corners=True)
         x_sub4 = self.res_block3_identity(x_sub4)
 
         x_sub4 = self.res_block4(x_sub4)
@@ -209,18 +209,19 @@ def forward(self, x):
         x_sub24, sub4_cls = self.cff_sub24(x_sub4, x_sub2)
         x_sub12, sub24_cls = self.cff_sub12(x_sub24, x_sub1)
 
-        x_sub12 = F.upsample(
-            x_sub12, size=get_interp_size(x_sub12, z_factor=2), mode="bilinear"
+        x_sub12 = F.interpolate(
+            x_sub12, size=get_interp_size(x_sub12, z_factor=2), mode="bilinear", align_corners=True
         )
         sub124_cls = self.classification(x_sub12)
 
         if self.training:
-            return sub4_cls, sub24_cls, sub124_cls
+            return (sub124_cls, sub24_cls, sub4_cls)
         else:  # eval mode
-            sub124_cls = F.upsample(
+            sub124_cls = F.interpolate(
                 sub124_cls,
                 size=get_interp_size(sub124_cls, z_factor=4),
                 mode="bilinear",
+                align_corners=True
             )  # Test only
             return sub124_cls
 

ptsemseg/models/utils.py

@@ -3,8 +3,6 @@
 import numpy as np
 import torch.nn.functional as F
 
-from torch.autograd import Variable
-
 
 class conv2DBatchNorm(nn.Module):
     def __init__(
@@ -572,7 +570,7 @@ def forward(self, x):
                 # out = F.adaptive_avg_pool2d(x, output_size=(pool_size, pool_size))
                 if self.model_name != "icnet":
                     out = module(out)
-                out = F.upsample(out, size=(h, w), mode="bilinear")
+                out = F.interpolate(out, size=(h, w), mode="bilinear", align_corners=True)
                 output_slices.append(out)
 
             return torch.cat(output_slices, dim=1)
@@ -586,7 +584,7 @@ def forward(self, x):
                 # out = F.adaptive_avg_pool2d(x, output_size=(pool_size, pool_size))
                 if self.model_name != "icnet":
                     out = module(out)
-                out = F.upsample(out, size=(h, w), mode="bilinear")
+                out = F.interpolate(out, size=(h, w), mode="bilinear", align_corners=True)
                 pp_sum = pp_sum + out
 
             return pp_sum
@@ -791,8 +789,8 @@ def __init__(
         )
 
     def forward(self, x_low, x_high):
-        x_low_upsampled = F.upsample(
-            x_low, size=get_interp_size(x_low, z_factor=2), mode="bilinear"
+        x_low_upsampled = F.interpolate(
+            x_low, size=get_interp_size(x_low, z_factor=2), mode="bilinear", align_corners=True
         )
 
         low_cls = self.low_classifier_conv(x_low_upsampled)
@@ -824,16 +822,13 @@ def interp(input, output_size, mode="bilinear"):
     oh, ow = output_size
 
     # normalize to [-1, 1]
-    h = torch.arange(0, oh) / (oh - 1) * 2 - 1
-    w = torch.arange(0, ow) / (ow - 1) * 2 - 1
+    h = torch.arange(0, oh, dtype=torch.float, device='cuda' if input.is_cuda else 'cpu') / (oh - 1) * 2 - 1
+    w = torch.arange(0, ow, dtype=torch.float, device='cuda' if input.is_cuda else 'cpu') / (ow - 1) * 2 - 1
 
-    grid = torch.zeros(oh, ow, 2)
+    grid = torch.zeros(oh, ow, 2, dtype=torch.float, device='cuda' if input.is_cuda else 'cpu')
     grid[:, :, 0] = w.unsqueeze(0).repeat(oh, 1)
     grid[:, :, 1] = h.unsqueeze(0).repeat(ow, 1).transpose(0, 1)
     grid = grid.unsqueeze(0).repeat(n, 1, 1, 1)  # grid.shape: [n, oh, ow, 2]
-    grid = Variable(grid)
-    if input.is_cuda:
-        grid = grid.cuda()
 
     return F.grid_sample(input, grid, mode=mode)
 

validate.py

@@ -36,7 +36,7 @@ def validate(cfg, args):
         split=cfg['data']['val_split'],
         is_transform=True,
         img_size=(cfg['data']['img_rows'], 
-                  cfg['data']['img_rows']),
+                  cfg['data']['img_cols']),
     )
 
     n_classes = loader.n_classes