support python2

lib/modeling/FPN.py

@@ -78,7 +78,7 @@ class fpn(nn.Module):
     similarly for spatial_scale: e.g [1/32, 1/16, 1/8, 1/4]
     """
     def __init__(self, conv_body_func, fpn_level_info, P2only=False):
-        super().__init__()
+        super(fpn, self).__init__()
         self.fpn_level_info = fpn_level_info
         self.P2only = P2only
 
@@ -261,7 +261,7 @@ def forward(self, x):
 class topdown_lateral_module(nn.Module):
     """Add a top-down lateral module."""
     def __init__(self, dim_in_top, dim_in_lateral):
-        super().__init__()
+        super(topdown_lateral_module, self).__init__()
         self.dim_in_top = dim_in_top
         self.dim_in_lateral = dim_in_lateral
         self.dim_out = dim_in_top
@@ -324,7 +324,7 @@ def get_min_max_levels():
 class fpn_rpn_outputs(nn.Module):
     """Add RPN on FPN specific outputs."""
     def __init__(self, dim_in, spatial_scales):
-        super().__init__()
+        super(fpn_rpn_outputs, self).__init__()
         self.dim_in = dim_in
         self.spatial_scales = spatial_scales
         self.dim_out = self.dim_in

lib/modeling/ResNet.py

@@ -41,7 +41,7 @@ def ResNet152_conv5_body():
 
 class ResNet_convX_body(nn.Module):
     def __init__(self, block_counts):
-        super().__init__()
+        super(ResNet_convX_body, self).__init__()
         self.block_counts = block_counts
         self.convX = len(block_counts) + 1
         self.num_layers = (sum(block_counts) + 3 * (self.convX == 4)) * 3 + 2
@@ -59,9 +59,9 @@ def __init__(self, block_counts):
             stride_init = 2 if cfg.RESNETS.RES5_DILATION == 1 else 1
             self.res5, dim_in = add_stage(dim_in, 2048, dim_bottleneck * 8, block_counts[3],
                                           cfg.RESNETS.RES5_DILATION, stride_init)
-            self.spatial_scale = 1 / 32 * cfg.RESNETS.RES5_DILATION
+            self.spatial_scale = 1. / 32 * cfg.RESNETS.RES5_DILATION
         else:
-            self.spatial_scale = 1 / 16  # final feature scale wrt. original image scale
+            self.spatial_scale = 1. / 16  # final feature scale wrt. original image scale
 
         self.dim_out = dim_in
 
@@ -117,7 +117,7 @@ def forward(self, x):
 
 class ResNet_roi_conv5_head(nn.Module):
     def __init__(self, dim_in, roi_xform_func, spatial_scale):
-        super().__init__()
+        super(ResNet_roi_conv5_head, self).__init__()
         self.roi_xform = roi_xform_func
         self.spatial_scale = spatial_scale
 
@@ -248,7 +248,7 @@ class bottleneck_transformation(nn.Module):
 
     def __init__(self, inplanes, outplanes, innerplanes, stride=1, dilation=1, group=1,
                  downsample=None):
-        super().__init__()
+        super(bottleneck_transformation, self).__init__()
         # In original resnet, stride=2 is on 1x1.
         # In fb.torch resnet, stride=2 is on 3x3.
         (str1x1, str3x3) = (stride, 1) if cfg.RESNETS.STRIDE_1X1 else (1, stride)
@@ -298,7 +298,7 @@ class bottleneck_gn_transformation(nn.Module):
 
     def __init__(self, inplanes, outplanes, innerplanes, stride=1, dilation=1, group=1,
                  downsample=None):
-        super().__init__()
+        super(bottleneck_gn_transformation, self).__init__()
         # In original resnet, stride=2 is on 1x1.
         # In fb.torch resnet, stride=2 is on 3x3.
         (str1x1, str3x3) = (stride, 1) if cfg.RESNETS.STRIDE_1X1 else (1, stride)

lib/modeling/collect_and_distribute_fpn_rpn_proposals.py

@@ -37,7 +37,7 @@ class CollectAndDistributeFpnRpnProposalsOp(nn.Module):
         [labels, bbox_targets, bbox_inside_weights, bbox_outside_weights].
     """
     def __init__(self):
-        super().__init__()
+        super(CollectAndDistributeFpnRpnProposalsOp, self).__init__()
 
     def forward(self, inputs, roidb, im_info):
         """

lib/modeling/fast_rcnn_heads.py

@@ -11,7 +11,7 @@
 
 class fast_rcnn_outputs(nn.Module):
     def __init__(self, dim_in):
-        super().__init__()
+        super(fast_rcnn_outputs, self).__init__()
         self.cls_score = nn.Linear(dim_in, cfg.MODEL.NUM_CLASSES)
         if cfg.MODEL.CLS_AGNOSTIC_BBOX_REG:  # bg and fg
             self.bbox_pred = nn.Linear(dim_in, 4 * 2)
@@ -73,7 +73,7 @@ def fast_rcnn_losses(cls_score, bbox_pred, label_int32, bbox_targets,
 class roi_2mlp_head(nn.Module):
     """Add a ReLU MLP with two hidden layers."""
     def __init__(self, dim_in, roi_xform_func, spatial_scale):
-        super().__init__()
+        super(roi_2mlp_head, self).__init__()
         self.dim_in = dim_in
         self.roi_xform = roi_xform_func
         self.spatial_scale = spatial_scale
@@ -119,7 +119,7 @@ def forward(self, x, rpn_ret):
 class roi_Xconv1fc_head(nn.Module):
     """Add a X conv + 1fc head, as a reference if not using GroupNorm"""
     def __init__(self, dim_in, roi_xform_func, spatial_scale):
-        super().__init__()
+        super(roi_Xconv1fc_head, self).__init__()
         self.dim_in = dim_in
         self.roi_xform = roi_xform_func
         self.spatial_scale = spatial_scale
@@ -181,7 +181,7 @@ def forward(self, x, rpn_ret):
 class roi_Xconv1fc_gn_head(nn.Module):
     """Add a X conv + 1fc head, with GroupNorm"""
     def __init__(self, dim_in, roi_xform_func, spatial_scale):
-        super().__init__()
+        super(roi_Xconv1fc_gn_head, self).__init__()
         self.dim_in = dim_in
         self.roi_xform = roi_xform_func
         self.spatial_scale = spatial_scale

lib/modeling/generate_proposal_labels.py

@@ -7,7 +7,7 @@
 
 class GenerateProposalLabelsOp(nn.Module):
     def __init__(self):
-        super().__init__()
+        super(GenerateProposalLabelsOp, self).__init__()
 
     def forward(self, rpn_rois, roidb, im_info):
         """Op for generating training labels for RPN proposals. This is used

lib/modeling/generate_proposals.py

@@ -11,7 +11,7 @@
 
 class GenerateProposalsOp(nn.Module):
     def __init__(self, anchors, spatial_scale):
-        super().__init__()
+        super(GenerateProposalsOp, self).__init__()
         self._anchors = anchors
         self._num_anchors = self._anchors.shape[0]
         self._feat_stride = 1. / spatial_scale

lib/modeling/keypoint_rcnn_heads.py

@@ -17,7 +17,7 @@
 class keypoint_outputs(nn.Module):
     """Mask R-CNN keypoint specific outputs: keypoint heatmaps."""
     def __init__(self, dim_in):
-        super().__init__()
+        super(keypoint_outputs, self).__init__()
         self.upsample_heatmap = (cfg.KRCNN.UP_SCALE > 1)
 
         if cfg.KRCNN.USE_DECONV:
@@ -129,7 +129,7 @@ def keypoint_losses(kps_pred, keypoint_locations_int32, keypoint_weights,
 class roi_pose_head_v1convX(nn.Module):
     """Mask R-CNN keypoint head. v1convX design: X * (conv)."""
     def __init__(self, dim_in, roi_xform_func, spatial_scale):
-        super().__init__()
+        super(roi_pose_head_v1convX, self).__init__()
         self.dim_in = dim_in
         self.roi_xform = roi_xform_func
         self.spatial_scale = spatial_scale

lib/modeling/mask_rcnn_heads.py

@@ -20,7 +20,7 @@
 class mask_rcnn_outputs(nn.Module):
     """Mask R-CNN specific outputs: either mask logits or probs."""
     def __init__(self, dim_in):
-        super().__init__()
+        super(mask_rcnn_outputs, self).__init__()
         self.dim_in = dim_in
 
         n_classes = cfg.MODEL.NUM_CLASSES if cfg.MRCNN.CLS_SPECIFIC_MASK else 1
@@ -127,7 +127,7 @@ def mask_rcnn_fcn_head_v1up(dim_in, roi_xform_func, spatial_scale):
 class mask_rcnn_fcn_head_v1upXconvs(nn.Module):
     """v1upXconvs design: X * (conv 3x3), convT 2x2."""
     def __init__(self, dim_in, roi_xform_func, spatial_scale, num_convs):
-        super().__init__()
+        super(mask_rcnn_fcn_head_v1upXconvs, self).__init__()
         self.dim_in = dim_in
         self.roi_xform = roi_xform_func
         self.spatial_scale = spatial_scale
@@ -191,7 +191,7 @@ def forward(self, x, rpn_ret):
 class mask_rcnn_fcn_head_v1upXconvs_gn(nn.Module):
     """v1upXconvs design: X * (conv 3x3), convT 2x2, with GroupNorm"""
     def __init__(self, dim_in, roi_xform_func, spatial_scale, num_convs):
-        super().__init__()
+        super(mask_rcnn_fcn_head_v1upXconvs_gn, self).__init__()
         self.dim_in = dim_in
         self.roi_xform = roi_xform_func
         self.spatial_scale = spatial_scale
@@ -263,7 +263,7 @@ class mask_rcnn_fcn_head_v0upshare(nn.Module):
     v0upshare design: conv5, convT 2x2.
     """
     def __init__(self, dim_in, roi_xform_func, spatial_scale):
-        super().__init__()
+        super(mask_rcnn_fcn_head_v0upshare, self).__init__()
         self.dim_in = dim_in
         self.roi_xform = roi_xform_func
         self.spatial_scale = spatial_scale
@@ -330,7 +330,7 @@ def forward(self, x, rpn_ret, roi_has_mask_int32=None):
 class mask_rcnn_fcn_head_v0up(nn.Module):
     """v0up design: conv5, deconv 2x2 (no weight sharing with the box head)."""
     def __init__(self, dim_in, roi_xform_func, spatial_scale):
-        super().__init__()
+        super(mask_rcnn_fcn_head_v0up, self).__init__()
         self.dim_in = dim_in
         self.roi_xform = roi_xform_func
         self.spatial_scale = spatial_scale

lib/modeling/model_builder.py

@@ -70,7 +70,7 @@ def wrapper(self, *args, **kwargs):
 
 class Generalized_RCNN(nn.Module):
     def __init__(self):
-        super().__init__()
+        super(Generalized_RCNN, self).__init__()
 
         # For cache
         self.mapping_to_detectron = None

lib/modeling/rpn_heads.py

@@ -37,7 +37,7 @@ def generic_rpn_losses(*inputs, **kwargs):
 class single_scale_rpn_outputs(nn.Module):
     """Add RPN outputs to a single scale model (i.e., no FPN)."""
     def __init__(self, dim_in, spatial_scale):
-        super().__init__()
+        super(single_scale_rpn_outputs, self).__init__()
         self.dim_in = dim_in
         self.dim_out = dim_in if cfg.RPN.OUT_DIM_AS_IN_DIM else cfg.RPN.OUT_DIM
         anchors = generate_anchors(

lib/nn/modules/affine.py

@@ -5,7 +5,7 @@
 class AffineChannel2d(nn.Module):
     """ A simple channel-wise affine transformation operation """
     def __init__(self, num_features):
-        super().__init__()
+        super(AffineChannel2d, self).__init__()
         self.num_features = num_features
         self.weight = nn.Parameter(torch.Tensor(num_features))
         self.bias = nn.Parameter(torch.Tensor(num_features))

lib/nn/modules/normalization.py

@@ -8,7 +8,7 @@
 
 class GroupNorm(nn.Module):
     def __init__(self, num_groups, num_channels, eps=1e-5, affine=True):
-        super().__init__()
+        super(GroupNorm, self).__init__()
         self.num_groups = num_groups
         self.num_channels = num_channels
         self.eps = eps

lib/nn/modules/upsample.py

@@ -15,7 +15,7 @@ class BilinearInterpolation2d(nn.Module):
     See: https://github.com/shelhamer/fcn.berkeleyvision.org/blob/master/surgery.py
     """
     def __init__(self, in_channels, out_channels, up_scale):
-        super().__init__()
+        super(BilinearInterpolation2d, self).__init__()
         assert in_channels == out_channels
         assert up_scale % 2 == 0, 'Scale should be even'
         self.in_channes = in_channels