CBNet modification

configs/cascade_rcnn_baselines/e2e_mask_cascade_rcnn_dual-X-152-32x8d-FPN-IN5k_1.44x.yaml

@@ -0,0 +1,88 @@
+MODEL:
+  TYPE: generalized_rcnn
+  CONV_BODY: FPN.add_fpn_ResNet152_conv5_body
+  NUM_CLASSES: 81
+  FASTER_RCNN: True
+  MASK_ON: True
+  CASCADE_ON: True
+  CLS_AGNOSTIC_BBOX_REG: True  # default: False
+NUM_GPUS: 8
+SOLVER:
+  WEIGHT_DECAY: 0.0001
+  LR_POLICY: steps_with_decay
+  # 1.44x schedule (note TRAIN.IMS_PER_BATCH: 1)
+  BASE_LR: 0.005
+  GAMMA: 0.1
+  MAX_ITER: 260000
+  STEPS: [0, 200000, 240000]
+FPN:
+  FPN_ON: True
+  MULTILEVEL_ROIS: True
+  MULTILEVEL_RPN: True
+RESNETS:
+  STRIDE_1X1: False  # default True for MSRA; False for C2 or Torch models
+  TRANS_FUNC: bottleneck_transformation
+  NUM_GROUPS: 32
+  WIDTH_PER_GROUP: 8
+FAST_RCNN:
+  ROI_BOX_HEAD: fast_rcnn_heads.add_roi_2mlp_head
+  ROI_XFORM_METHOD: RoIAlign
+  ROI_XFORM_RESOLUTION: 7
+  ROI_XFORM_SAMPLING_RATIO: 2
+MRCNN:
+  ROI_MASK_HEAD: mask_rcnn_heads.mask_rcnn_fcn_head_v1up4convs
+  RESOLUTION: 28  # (output mask resolution) default 14
+  ROI_XFORM_METHOD: RoIAlign
+  ROI_XFORM_RESOLUTION: 14  # default 7
+  ROI_XFORM_SAMPLING_RATIO: 2  # default 0
+  DILATION: 1  # default 2
+  CONV_INIT: MSRAFill  # default GaussianFill
+  AT_STAGE: 3
+CASCADE_RCNN:
+  ROI_BOX_HEAD: cascade_rcnn_heads.add_roi_2mlp_head
+  NUM_STAGE: 3
+  TEST_STAGE: 3
+  TEST_ENSEMBLE: True
+TRAIN:
+  WEIGHTS: https://s3-us-west-2.amazonaws.com/detectron/ImageNetPretrained/25093814/X-152-32x8d-IN5k.pkl
+  DATASETS: ('coco_2014_train', 'coco_2014_valminusminival')
+  SCALES: (640, 672, 704, 736, 768, 800)  # Scale jitter
+  MAX_SIZE: 1333
+  IMS_PER_BATCH: 1
+  BATCH_SIZE_PER_IM: 512
+  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
+TEST:
+  DATASETS: ('coco_2015_test-dev',)
+  #DATASETS: ('coco_2014_minival',)
+  SCALE: 800
+  MAX_SIZE: 1333
+  NMS: 0.5
+  #BBOX_VOTE:
+  #  ENABLED: True
+  #  VOTE_TH: 0.9
+  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level
+  RPN_POST_NMS_TOP_N: 1000
+  #BBOX_AUG:
+  #  ENABLED: True
+  #  SCORE_HEUR: UNION
+  #  COORD_HEUR: UNION
+  #  H_FLIP: True
+    #SCALES: (400, 500, 600, 700, 900, 1000, 1100, 1200)
+    #MAX_SIZE: 2000
+  #  SCALES: (400, 500, 600, 700, 900, 1000, 1100)
+  #  MAX_SIZE: 1500
+  #  SCALE_H_FLIP: False #True
+  #  SCALE_SIZE_DEP: False
+  #  ASPECT_RATIOS: ()
+  #  ASPECT_RATIO_H_FLIP: False
+  #MASK_AUG:
+  #  ENABLED: True
+  #  HEUR: SOFT_AVG
+  #  H_FLIP: True
+  #  SCALES: (400, 500, 600, 700, 900, 1000)
+  #  MAX_SIZE: 1500
+  #  SCALE_H_FLIP: True
+  #  SCALE_SIZE_DEP: False
+  #  ASPECT_RATIOS: ()
+  #  ASPECT_RATIO_H_FLIP: False
+OUTPUT_DIR: .

detectron/modeling/FPN.py

@@ -60,7 +60,8 @@ def add_fpn_ResNet50_conv5_P2only_body(model):
 
 def add_fpn_ResNet101_conv5_body(model):
     return add_fpn_onto_conv_body(
-        model, ResNet.add_ResNet101_conv5_body, fpn_level_info_ResNet101_conv5
+        model, ResNet.add_ResNet101_conv5_body, ResNet.add_ResNet101_conv5_body_old,
+        fpn_level_info_ResNet101_conv5, fpn_level_info_ResNet101_conv5_old
     )
 
 
@@ -72,12 +73,17 @@ def add_fpn_ResNet101_conv5_P2only_body(model):
         P2only=True
     )
 
+def add_fpn_ResNet152_conv5_body(model):
+    return add_fpn_onto_conv_body(
+        model, ResNet.add_ResNet152_conv5_body, ResNet.add_ResNet152_conv5_body_old, fpn_level_info_ResNet152_conv5,fpn_level_info_ResNet152_conv5_old
+    )
 
+'''
 def add_fpn_ResNet152_conv5_body(model):
     return add_fpn_onto_conv_body(
         model, ResNet.add_ResNet152_conv5_body, fpn_level_info_ResNet152_conv5
     )
-
+'''
 
 def add_fpn_ResNet152_conv5_P2only_body(model):
     return add_fpn_onto_conv_body(
@@ -93,15 +99,17 @@ def add_fpn_ResNet152_conv5_P2only_body(model):
 # ---------------------------------------------------------------------------- #
 
 def add_fpn_onto_conv_body(
-    model, conv_body_func, fpn_level_info_func, P2only=False
+    model, conv_body_func_now, conv_body_func_old, fpn_level_info_func,
+    fpn_level_info_func_old, P2only=False
 ):
     """Add the specified conv body to the model and then add FPN levels to it.
     """
     # Note: blobs_conv is in revsersed order: [fpn5, fpn4, fpn3, fpn2]
     # similarly for dims_conv: [2048, 1024, 512, 256]
     # similarly for spatial_scales_fpn: [1/32, 1/16, 1/8, 1/4]
 
-    conv_body_func(model)
+    conv_body_func_old(model)
+    conv_body_func_now(model)
     blobs_fpn, dim_fpn, spatial_scales_fpn = add_fpn(
         model, fpn_level_info_func()
     )
@@ -278,6 +286,38 @@ def add_topdown_lateral_module(
             bias_init=const_fill(0.0)
         )
     else:
+        '''
+        tmp1 = model.Conv(#########reduce channel /2
+            fpn_lataral_old,
+            fpn_bottom + '_mid1',
+            dim_in = dim_lateral * 2,
+            dim_out = dim_lateral,
+            kernel=1,
+            pad=0,
+            stride=1,
+            weight_init=(
+                const_fill(0.0)
+                if cfg.FPN.ZERO_INIT_LATERAL else ('XavierFill', {})
+            ),
+            bias_init=const_fill(0.0)
+        )
+        tmp2 = model.net.UpsampleNearest(tmp1, fpn_bottom + '_mid2', scale=2)##upsample
+        tmp3 = model.net.Sum([tmp2, fpn_lateral], fpn_bottom + '_mid3')##### add
+        lat = model.Conv(
+            tmp3,
+            fpn_bottom + '_lateral',
+            dim_in=dim_lateral,
+            dim_out=dim_top,
+            kernel=1,
+            pad=0,
+            stride=1,
+            weight_init=(
+                const_fill(0.0)
+                if cfg.FPN.ZERO_INIT_LATERAL else ('XavierFill', {})
+            ),
+            bias_init=const_fill(0.0)
+        )
+        '''
         lat = model.Conv(
             fpn_lateral,
             fpn_bottom + '_lateral',
@@ -292,6 +332,7 @@ def add_topdown_lateral_module(
             ),
             bias_init=const_fill(0.0)
         )
+
     # Top-down 2x upsampling
     td = model.net.UpsampleNearest(fpn_top, fpn_bottom + '_topdown', scale=2)
     # Sum lateral and top-down
@@ -559,10 +600,25 @@ def fpn_level_info_ResNet101_conv5():
         spatial_scales=(1. / 32., 1. / 16., 1. / 8., 1. / 4.)
     )
 
+def fpn_level_info_ResNet101_conv5_old():
+    return FpnLevelInfo(
+        blobs=('old_res5_2_sum', 'old_res4_22_sum', 'old_res3_3_sum', 'old_res2_2_sum'),
+        dims=(2048, 1024, 512, 256),
+        spatial_scales=(1. / 32., 1. / 16., 1. / 8., 1. / 4.)
+    )
+
 
 def fpn_level_info_ResNet152_conv5():
     return FpnLevelInfo(
         blobs=('res5_2_sum', 'res4_35_sum', 'res3_7_sum', 'res2_2_sum'),
         dims=(2048, 1024, 512, 256),
         spatial_scales=(1. / 32., 1. / 16., 1. / 8., 1. / 4.)
     )
+
+def fpn_level_info_ResNet152_conv5_old():
+    return FpnLevelInfo(
+        blobs=('old_res5_2_sum', 'old_res4_35_sum', 'old_res3_7_sum', 'old_res2_2_sum'),
+        dims=(2048, 1024, 512, 256),
+        spatial_scales=(1. / 32., 1. / 16., 1. / 8., 1. / 4.)
+    )
+