Remove legacy multi-scale inference support

Reviewed By: ir413 Differential Revision: D7148425 fbshipit-source-id: 78c11427b0559248116ca1065b8252355f0034a3

Remove legacy multi-scale inference support
Reviewed By: ir413 Differential Revision: D7148425 fbshipit-source-id: 78c11427b0559248116ca1065b8252355f0034a3
70e20023 · Ross Girshick · Facebook Github Bot · 9b41e84b · 70e20023 · 70e20023
Commit 70e20023 authored Mar 20, 2018 by Ross Girshick Committed by Facebook Github Bot Mar 20, 2018
87 changed files
--- a/configs/12_2017_baselines/e2e_faster_rcnn_R-101-FPN_1x.yaml
+++ b/configs/12_2017_baselines/e2e_faster_rcnn_R-101-FPN_1x.yaml
@@ -29,7 +29,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_faster_rcnn_R-101-FPN_2x.yaml
+++ b/configs/12_2017_baselines/e2e_faster_rcnn_R-101-FPN_2x.yaml
@@ -29,7 +29,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_faster_rcnn_R-50-C4_1x.yaml
+++ b/configs/12_2017_baselines/e2e_faster_rcnn_R-50-C4_1x.yaml
@@ -26,7 +26,7 @@ TRAIN:
  BATCH_SIZE_PER_IM: 512
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 6000

--- a/configs/12_2017_baselines/e2e_faster_rcnn_R-50-C4_2x.yaml
+++ b/configs/12_2017_baselines/e2e_faster_rcnn_R-50-C4_2x.yaml
@@ -26,7 +26,7 @@ TRAIN:
  BATCH_SIZE_PER_IM: 512
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 6000

--- a/configs/12_2017_baselines/e2e_faster_rcnn_R-50-FPN_1x.yaml
+++ b/configs/12_2017_baselines/e2e_faster_rcnn_R-50-FPN_1x.yaml
@@ -29,7 +29,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_faster_rcnn_R-50-FPN_2x.yaml
+++ b/configs/12_2017_baselines/e2e_faster_rcnn_R-50-FPN_2x.yaml
@@ -29,7 +29,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_faster_rcnn_X-101-32x8d-FPN_1x.yaml
+++ b/configs/12_2017_baselines/e2e_faster_rcnn_X-101-32x8d-FPN_1x.yaml
@@ -36,7 +36,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_faster_rcnn_X-101-32x8d-FPN_2x.yaml
+++ b/configs/12_2017_baselines/e2e_faster_rcnn_X-101-32x8d-FPN_2x.yaml
@@ -36,7 +36,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_faster_rcnn_X-101-64x4d-FPN_1x.yaml
+++ b/configs/12_2017_baselines/e2e_faster_rcnn_X-101-64x4d-FPN_1x.yaml
@@ -36,7 +36,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_faster_rcnn_X-101-64x4d-FPN_2x.yaml
+++ b/configs/12_2017_baselines/e2e_faster_rcnn_X-101-64x4d-FPN_2x.yaml
@@ -37,7 +37,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_keypoint_rcnn_R-101-FPN_1x.yaml
+++ b/configs/12_2017_baselines/e2e_keypoint_rcnn_R-101-FPN_1x.yaml
@@ -43,7 +43,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('keypoints_coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_keypoint_rcnn_R-101-FPN_s1x.yaml
+++ b/configs/12_2017_baselines/e2e_keypoint_rcnn_R-101-FPN_s1x.yaml
@@ -43,7 +43,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('keypoints_coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_keypoint_rcnn_R-50-FPN_1x.yaml
+++ b/configs/12_2017_baselines/e2e_keypoint_rcnn_R-50-FPN_1x.yaml
@@ -43,7 +43,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('keypoints_coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_keypoint_rcnn_R-50-FPN_s1x.yaml
+++ b/configs/12_2017_baselines/e2e_keypoint_rcnn_R-50-FPN_s1x.yaml
@@ -43,7 +43,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('keypoints_coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_keypoint_rcnn_X-101-32x8d-FPN_1x.yaml
+++ b/configs/12_2017_baselines/e2e_keypoint_rcnn_X-101-32x8d-FPN_1x.yaml
@@ -48,7 +48,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('keypoints_coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_keypoint_rcnn_X-101-32x8d-FPN_s1x.yaml
+++ b/configs/12_2017_baselines/e2e_keypoint_rcnn_X-101-32x8d-FPN_s1x.yaml
@@ -48,7 +48,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('keypoints_coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_keypoint_rcnn_X-101-64x4d-FPN_1x.yaml
+++ b/configs/12_2017_baselines/e2e_keypoint_rcnn_X-101-64x4d-FPN_1x.yaml
@@ -49,7 +49,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('keypoints_coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_keypoint_rcnn_X-101-64x4d-FPN_s1x.yaml
+++ b/configs/12_2017_baselines/e2e_keypoint_rcnn_X-101-64x4d-FPN_s1x.yaml
@@ -49,7 +49,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('keypoints_coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_mask_rcnn_R-101-FPN_1x.yaml
+++ b/configs/12_2017_baselines/e2e_mask_rcnn_R-101-FPN_1x.yaml
@@ -38,7 +38,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_mask_rcnn_R-101-FPN_2x.yaml
+++ b/configs/12_2017_baselines/e2e_mask_rcnn_R-101-FPN_2x.yaml
@@ -38,7 +38,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_mask_rcnn_R-50-C4_1x.yaml
+++ b/configs/12_2017_baselines/e2e_mask_rcnn_R-50-C4_1x.yaml
@@ -34,7 +34,7 @@ TRAIN:
  BATCH_SIZE_PER_IM: 512
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 6000

--- a/configs/12_2017_baselines/e2e_mask_rcnn_R-50-C4_2x.yaml
+++ b/configs/12_2017_baselines/e2e_mask_rcnn_R-50-C4_2x.yaml
@@ -34,7 +34,7 @@ TRAIN:
  BATCH_SIZE_PER_IM: 512
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 6000

--- a/configs/12_2017_baselines/e2e_mask_rcnn_R-50-FPN_1x.yaml
+++ b/configs/12_2017_baselines/e2e_mask_rcnn_R-50-FPN_1x.yaml
@@ -38,7 +38,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_mask_rcnn_R-50-FPN_2x.yaml
+++ b/configs/12_2017_baselines/e2e_mask_rcnn_R-50-FPN_2x.yaml
@@ -38,7 +38,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_mask_rcnn_X-101-32x8d-FPN_1x.yaml
+++ b/configs/12_2017_baselines/e2e_mask_rcnn_X-101-32x8d-FPN_1x.yaml
@@ -45,7 +45,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_mask_rcnn_X-101-32x8d-FPN_2x.yaml
+++ b/configs/12_2017_baselines/e2e_mask_rcnn_X-101-32x8d-FPN_2x.yaml
@@ -45,7 +45,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_mask_rcnn_X-101-64x4d-FPN_1x.yaml
+++ b/configs/12_2017_baselines/e2e_mask_rcnn_X-101-64x4d-FPN_1x.yaml
@@ -46,7 +46,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_mask_rcnn_X-101-64x4d-FPN_2x.yaml
+++ b/configs/12_2017_baselines/e2e_mask_rcnn_X-101-64x4d-FPN_2x.yaml
@@ -46,7 +46,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/12_2017_baselines/e2e_mask_rcnn_X-152-32x8d-FPN-IN5k_1.44x.yaml
+++ b/configs/12_2017_baselines/e2e_mask_rcnn_X-152-32x8d-FPN-IN5k_1.44x.yaml
@@ -45,7 +45,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  BBOX_VOTE:

--- a/configs/12_2017_baselines/fast_rcnn_R-101-FPN_1x.yaml
+++ b/configs/12_2017_baselines/fast_rcnn_R-101-FPN_1x.yaml
@@ -30,7 +30,7 @@ TEST:
  DATASETS: ('coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/35998887/12_2017_baselines/rpn_R-101-FPN_1x.yaml.08_07_07.vzhHEs0V/output/test/coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/fast_rcnn_R-101-FPN_2x.yaml
+++ b/configs/12_2017_baselines/fast_rcnn_R-101-FPN_2x.yaml
@@ -30,7 +30,7 @@ TEST:
  DATASETS: ('coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/35998887/12_2017_baselines/rpn_R-101-FPN_1x.yaml.08_07_07.vzhHEs0V/output/test/coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/fast_rcnn_R-50-C4_1x.yaml
+++ b/configs/12_2017_baselines/fast_rcnn_R-50-C4_1x.yaml
@@ -28,7 +28,7 @@ TEST:
  DATASETS: ('coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/35998355/12_2017_baselines/rpn_R-50-C4_1x.yaml.08_00_43.njH5oD9L/output/test/coco_2014_minival/rpn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/fast_rcnn_R-50-C4_2x.yaml
+++ b/configs/12_2017_baselines/fast_rcnn_R-50-C4_2x.yaml
@@ -28,7 +28,7 @@ TEST:
  DATASETS: ('coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/35998355/12_2017_baselines/rpn_R-50-C4_1x.yaml.08_00_43.njH5oD9L/output/test/coco_2014_minival/rpn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/fast_rcnn_R-50-FPN_1x.yaml
+++ b/configs/12_2017_baselines/fast_rcnn_R-50-FPN_1x.yaml
@@ -30,7 +30,7 @@ TEST:
  DATASETS: ('coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/35998814/12_2017_baselines/rpn_R-50-FPN_1x.yaml.08_06_03.Axg0r179/output/test/coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/fast_rcnn_R-50-FPN_2x.yaml
+++ b/configs/12_2017_baselines/fast_rcnn_R-50-FPN_2x.yaml
@@ -30,7 +30,7 @@ TEST:
  DATASETS: ('coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/35998814/12_2017_baselines/rpn_R-50-FPN_1x.yaml.08_06_03.Axg0r179/output/test/coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/fast_rcnn_X-101-32x8d-FPN_1x.yaml
+++ b/configs/12_2017_baselines/fast_rcnn_X-101-32x8d-FPN_1x.yaml
@@ -37,7 +37,7 @@ TEST:
  DATASETS: ('coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/36760102/12_2017_baselines/rpn_X-101-32x8d-FPN_1x.yaml.06_00_16.RWeBAniO/output/test/coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/fast_rcnn_X-101-32x8d-FPN_2x.yaml
+++ b/configs/12_2017_baselines/fast_rcnn_X-101-32x8d-FPN_2x.yaml
@@ -37,7 +37,7 @@ TEST:
  DATASETS: ('coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/36760102/12_2017_baselines/rpn_X-101-32x8d-FPN_1x.yaml.06_00_16.RWeBAniO/output/test/coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/fast_rcnn_X-101-64x4d-FPN_1x.yaml
+++ b/configs/12_2017_baselines/fast_rcnn_X-101-64x4d-FPN_1x.yaml
@@ -37,7 +37,7 @@ TEST:
  DATASETS: ('coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/35998956/12_2017_baselines/rpn_X-101-64x4d-FPN_1x.yaml.08_08_41.Seh0psKz/output/test/coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/fast_rcnn_X-101-64x4d-FPN_2x.yaml
+++ b/configs/12_2017_baselines/fast_rcnn_X-101-64x4d-FPN_2x.yaml
@@ -37,7 +37,7 @@ TEST:
  DATASETS: ('coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/35998956/12_2017_baselines/rpn_X-101-64x4d-FPN_1x.yaml.08_08_41.Seh0psKz/output/test/coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/keypoint_rcnn_R-101-FPN_1x.yaml
+++ b/configs/12_2017_baselines/keypoint_rcnn_R-101-FPN_1x.yaml
@@ -44,7 +44,7 @@ TEST:
  DATASETS: ('keypoints_coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/35999521/12_2017_baselines/rpn_person_only_R-101-FPN_1x.yaml.08_20_33.1OkqMmqP/output/test/keypoints_coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/keypoint_rcnn_R-101-FPN_s1x.yaml
+++ b/configs/12_2017_baselines/keypoint_rcnn_R-101-FPN_s1x.yaml
@@ -44,7 +44,7 @@ TEST:
  DATASETS: ('keypoints_coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/35999521/12_2017_baselines/rpn_person_only_R-101-FPN_1x.yaml.08_20_33.1OkqMmqP/output/test/keypoints_coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/keypoint_rcnn_R-50-FPN_1x.yaml
+++ b/configs/12_2017_baselines/keypoint_rcnn_R-50-FPN_1x.yaml
@@ -44,7 +44,7 @@ TEST:
  DATASETS: ('keypoints_coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/35998996/12_2017_baselines/rpn_person_only_R-50-FPN_1x.yaml.08_10_08.0ZWmJm6F/output/test/keypoints_coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/keypoint_rcnn_R-50-FPN_s1x.yaml
+++ b/configs/12_2017_baselines/keypoint_rcnn_R-50-FPN_s1x.yaml
@@ -44,7 +44,7 @@ TEST:
  DATASETS: ('keypoints_coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/35998996/12_2017_baselines/rpn_person_only_R-50-FPN_1x.yaml.08_10_08.0ZWmJm6F/output/test/keypoints_coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/keypoint_rcnn_X-101-32x8d-FPN_1x.yaml
+++ b/configs/12_2017_baselines/keypoint_rcnn_X-101-32x8d-FPN_1x.yaml
@@ -49,7 +49,7 @@ TEST:
  DATASETS: ('keypoints_coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/36760438/12_2017_baselines/rpn_person_only_X-101-32x8d-FPN_1x.yaml.06_04_23.M2oJlDPW/output/test/keypoints_coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/keypoint_rcnn_X-101-32x8d-FPN_s1x.yaml
+++ b/configs/12_2017_baselines/keypoint_rcnn_X-101-32x8d-FPN_s1x.yaml
@@ -49,7 +49,7 @@ TEST:
  DATASETS: ('keypoints_coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/36760438/12_2017_baselines/rpn_person_only_X-101-32x8d-FPN_1x.yaml.06_04_23.M2oJlDPW/output/test/keypoints_coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/keypoint_rcnn_X-101-64x4d-FPN_1x.yaml
+++ b/configs/12_2017_baselines/keypoint_rcnn_X-101-64x4d-FPN_1x.yaml
@@ -50,7 +50,7 @@ TEST:
  DATASETS: ('keypoints_coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/35999553/12_2017_baselines/rpn_person_only_X-101-64x4d-FPN_1x.yaml.08_21_33.ghFzzArr/output/test/keypoints_coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/keypoint_rcnn_X-101-64x4d-FPN_s1x.yaml
+++ b/configs/12_2017_baselines/keypoint_rcnn_X-101-64x4d-FPN_s1x.yaml
@@ -50,7 +50,7 @@ TEST:
  DATASETS: ('keypoints_coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/35999553/12_2017_baselines/rpn_person_only_X-101-64x4d-FPN_1x.yaml.08_21_33.ghFzzArr/output/test/keypoints_coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/mask_rcnn_R-101-FPN_1x.yaml
+++ b/configs/12_2017_baselines/mask_rcnn_R-101-FPN_1x.yaml
@@ -39,7 +39,7 @@ TEST:
  DATASETS: ('coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/35998887/12_2017_baselines/rpn_R-101-FPN_1x.yaml.08_07_07.vzhHEs0V/output/test/coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/mask_rcnn_R-101-FPN_2x.yaml
+++ b/configs/12_2017_baselines/mask_rcnn_R-101-FPN_2x.yaml
@@ -39,7 +39,7 @@ TEST:
  DATASETS: ('coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/35998887/12_2017_baselines/rpn_R-101-FPN_1x.yaml.08_07_07.vzhHEs0V/output/test/coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/mask_rcnn_R-50-C4_1x.yaml
+++ b/configs/12_2017_baselines/mask_rcnn_R-50-C4_1x.yaml
@@ -36,7 +36,7 @@ TEST:
  DATASETS: ('coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/35998355/12_2017_baselines/rpn_R-50-C4_1x.yaml.08_00_43.njH5oD9L/output/test/coco_2014_minival/rpn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/mask_rcnn_R-50-C4_2x.yaml
+++ b/configs/12_2017_baselines/mask_rcnn_R-50-C4_2x.yaml
@@ -36,7 +36,7 @@ TEST:
  DATASETS: ('coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/35998355/12_2017_baselines/rpn_R-50-C4_1x.yaml.08_00_43.njH5oD9L/output/test/coco_2014_minival/rpn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/mask_rcnn_R-50-FPN_1x.yaml
+++ b/configs/12_2017_baselines/mask_rcnn_R-50-FPN_1x.yaml
@@ -39,7 +39,7 @@ TEST:
  DATASETS: ('coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/35998814/12_2017_baselines/rpn_R-50-FPN_1x.yaml.08_06_03.Axg0r179/output/test/coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/mask_rcnn_R-50-FPN_2x.yaml
+++ b/configs/12_2017_baselines/mask_rcnn_R-50-FPN_2x.yaml
@@ -39,7 +39,7 @@ TEST:
  DATASETS: ('coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/35998814/12_2017_baselines/rpn_R-50-FPN_1x.yaml.08_06_03.Axg0r179/output/test/coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/mask_rcnn_X-101-32x8d-FPN_1x.yaml
+++ b/configs/12_2017_baselines/mask_rcnn_X-101-32x8d-FPN_1x.yaml
@@ -46,7 +46,7 @@ TEST:
  DATASETS: ('coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/36760102/12_2017_baselines/rpn_X-101-32x8d-FPN_1x.yaml.06_00_16.RWeBAniO/output/test/coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/mask_rcnn_X-101-32x8d-FPN_2x.yaml
+++ b/configs/12_2017_baselines/mask_rcnn_X-101-32x8d-FPN_2x.yaml
@@ -46,7 +46,7 @@ TEST:
  DATASETS: ('coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/36760102/12_2017_baselines/rpn_X-101-32x8d-FPN_1x.yaml.06_00_16.RWeBAniO/output/test/coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/mask_rcnn_X-101-64x4d-FPN_1x.yaml
+++ b/configs/12_2017_baselines/mask_rcnn_X-101-64x4d-FPN_1x.yaml
@@ -47,7 +47,7 @@ TEST:
  DATASETS: ('coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/35998956/12_2017_baselines/rpn_X-101-64x4d-FPN_1x.yaml.08_08_41.Seh0psKz/output/test/coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/mask_rcnn_X-101-64x4d-FPN_2x.yaml
+++ b/configs/12_2017_baselines/mask_rcnn_X-101-64x4d-FPN_2x.yaml
@@ -47,7 +47,7 @@ TEST:
  DATASETS: ('coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/35998956/12_2017_baselines/rpn_X-101-64x4d-FPN_1x.yaml.08_08_41.Seh0psKz/output/test/coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/retinanet_R-101-FPN_1x.yaml
+++ b/configs/12_2017_baselines/retinanet_R-101-FPN_1x.yaml
@@ -33,7 +33,7 @@ TRAIN:
  RPN_STRADDLE_THRESH: -1  # default 0
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 10000  # Per FPN level

--- a/configs/12_2017_baselines/retinanet_R-101-FPN_2x.yaml
+++ b/configs/12_2017_baselines/retinanet_R-101-FPN_2x.yaml
@@ -33,7 +33,7 @@ TRAIN:
  RPN_STRADDLE_THRESH: -1  # default 0
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 10000  # Per FPN level

--- a/configs/12_2017_baselines/retinanet_R-50-FPN_1x.yaml
+++ b/configs/12_2017_baselines/retinanet_R-50-FPN_1x.yaml
@@ -33,7 +33,7 @@ TRAIN:
  RPN_STRADDLE_THRESH: -1  # default 0
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 10000  # Per FPN level

--- a/configs/12_2017_baselines/retinanet_R-50-FPN_2x.yaml
+++ b/configs/12_2017_baselines/retinanet_R-50-FPN_2x.yaml
@@ -33,7 +33,7 @@ TRAIN:
  RPN_STRADDLE_THRESH: -1  # default 0
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 10000  # Per FPN level

--- a/configs/12_2017_baselines/retinanet_X-101-32x8d-FPN_1x.yaml
+++ b/configs/12_2017_baselines/retinanet_X-101-32x8d-FPN_1x.yaml
@@ -38,7 +38,7 @@ TRAIN:
  RPN_STRADDLE_THRESH: -1  # default 0
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 10000  # Per FPN level

--- a/configs/12_2017_baselines/retinanet_X-101-32x8d-FPN_2x.yaml
+++ b/configs/12_2017_baselines/retinanet_X-101-32x8d-FPN_2x.yaml
@@ -38,7 +38,7 @@ TRAIN:
  RPN_STRADDLE_THRESH: -1  # default 0
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 10000  # Per FPN level

--- a/configs/12_2017_baselines/retinanet_X-101-64x4d-FPN_1x.yaml
+++ b/configs/12_2017_baselines/retinanet_X-101-64x4d-FPN_1x.yaml
@@ -38,7 +38,7 @@ TRAIN:
  RPN_STRADDLE_THRESH: -1  # default 0
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 10000  # Per FPN level

--- a/configs/12_2017_baselines/retinanet_X-101-64x4d-FPN_2x.yaml
+++ b/configs/12_2017_baselines/retinanet_X-101-64x4d-FPN_2x.yaml
@@ -38,7 +38,7 @@ TRAIN:
  RPN_STRADDLE_THRESH: -1  # default 0
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 10000  # Per FPN level

--- a/configs/12_2017_baselines/rpn_R-101-FPN_1x.yaml
+++ b/configs/12_2017_baselines/rpn_R-101-FPN_1x.yaml
@@ -25,7 +25,7 @@ TRAIN:
  MAX_SIZE: 1333
 TEST:
  DATASETS: ('coco_2014_minival','coco_2014_train','coco_2014_valminusminival')
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level
  RPN_POST_NMS_TOP_N: 2000

--- a/configs/12_2017_baselines/rpn_R-50-C4_1x.yaml
+++ b/configs/12_2017_baselines/rpn_R-50-C4_1x.yaml
@@ -20,7 +20,7 @@ TRAIN:
  MAX_SIZE: 1333
 TEST:
  DATASETS: ('coco_2014_minival','coco_2014_train','coco_2014_valminusminival')
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
 USE_NCCL: False
 OUTPUT_DIR: .
--- a/configs/12_2017_baselines/rpn_R-50-FPN_1x.yaml
+++ b/configs/12_2017_baselines/rpn_R-50-FPN_1x.yaml
@@ -25,7 +25,7 @@ TRAIN:
  MAX_SIZE: 1333
 TEST:
  DATASETS: ('coco_2014_minival','coco_2014_train','coco_2014_valminusminival')
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level
  RPN_POST_NMS_TOP_N: 2000

--- a/configs/12_2017_baselines/rpn_X-101-32x8d-FPN_1x.yaml
+++ b/configs/12_2017_baselines/rpn_X-101-32x8d-FPN_1x.yaml
@@ -30,7 +30,7 @@ TRAIN:
  MAX_SIZE: 1333
 TEST:
  DATASETS: ('coco_2014_minival','coco_2014_train','coco_2014_valminusminival')
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level
  RPN_POST_NMS_TOP_N: 2000

--- a/configs/12_2017_baselines/rpn_X-101-64x4d-FPN_1x.yaml
+++ b/configs/12_2017_baselines/rpn_X-101-64x4d-FPN_1x.yaml
@@ -30,7 +30,7 @@ TRAIN:
  MAX_SIZE: 1333
 TEST:
  DATASETS: ('coco_2014_minival','coco_2014_train','coco_2014_valminusminival')
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level
  RPN_POST_NMS_TOP_N: 2000

--- a/configs/12_2017_baselines/rpn_person_only_R-101-FPN_1x.yaml
+++ b/configs/12_2017_baselines/rpn_person_only_R-101-FPN_1x.yaml
@@ -25,7 +25,7 @@ TRAIN:
  MAX_SIZE: 1333
 TEST:
  DATASETS: ('keypoints_coco_2014_minival', 'keypoints_coco_2014_train', 'keypoints_coco_2014_valminusminival', 'keypoints_coco_2015_test')
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level
  RPN_POST_NMS_TOP_N: 2000

--- a/configs/12_2017_baselines/rpn_person_only_R-50-FPN_1x.yaml
+++ b/configs/12_2017_baselines/rpn_person_only_R-50-FPN_1x.yaml
@@ -25,7 +25,7 @@ TRAIN:
  MAX_SIZE: 1333
 TEST:
  DATASETS: ('keypoints_coco_2014_minival', 'keypoints_coco_2014_train', 'keypoints_coco_2014_valminusminival', 'keypoints_coco_2015_test')
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level
  RPN_POST_NMS_TOP_N: 2000

--- a/configs/12_2017_baselines/rpn_person_only_X-101-32x8d-FPN_1x.yaml
+++ b/configs/12_2017_baselines/rpn_person_only_X-101-32x8d-FPN_1x.yaml
@@ -30,7 +30,7 @@ TRAIN:
  MAX_SIZE: 1333
 TEST:
  DATASETS: ('keypoints_coco_2014_minival', 'keypoints_coco_2014_train', 'keypoints_coco_2014_valminusminival', 'keypoints_coco_2015_test')
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level
  RPN_POST_NMS_TOP_N: 2000

--- a/configs/12_2017_baselines/rpn_person_only_X-101-64x4d-FPN_1x.yaml
+++ b/configs/12_2017_baselines/rpn_person_only_X-101-64x4d-FPN_1x.yaml
@@ -30,7 +30,7 @@ TRAIN:
  MAX_SIZE: 1333
 TEST:
  DATASETS: ('keypoints_coco_2014_minival', 'keypoints_coco_2014_train', 'keypoints_coco_2014_valminusminival', 'keypoints_coco_2015_test')
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level
  RPN_POST_NMS_TOP_N: 2000

--- a/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml
+++ b/configs/getting_started/tutorial_1gpu_e2e_faster_rcnn_R-50-FPN.yaml
@@ -46,7 +46,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (500,)
+  SCALE: 500
  MAX_SIZE: 833
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/getting_started/tutorial_2gpu_e2e_faster_rcnn_R-50-FPN.yaml
+++ b/configs/getting_started/tutorial_2gpu_e2e_faster_rcnn_R-50-FPN.yaml
@@ -46,7 +46,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (500,)
+  SCALE: 500
  MAX_SIZE: 833
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/getting_started/tutorial_4gpu_e2e_faster_rcnn_R-50-FPN.yaml
+++ b/configs/getting_started/tutorial_4gpu_e2e_faster_rcnn_R-50-FPN.yaml
@@ -46,7 +46,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (500,)
+  SCALE: 500
  MAX_SIZE: 833
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/getting_started/tutorial_8gpu_e2e_faster_rcnn_R-50-FPN.yaml
+++ b/configs/getting_started/tutorial_8gpu_e2e_faster_rcnn_R-50-FPN.yaml
@@ -46,7 +46,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (500,)
+  SCALE: 500
  MAX_SIZE: 833
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/test_time_aug/e2e_mask_rcnn_R-50-FPN_2x.yaml
+++ b/configs/test_time_aug/e2e_mask_rcnn_R-50-FPN_2x.yaml
@@ -38,7 +38,7 @@ TRAIN:
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
 TEST:
  DATASETS: ('coco_2014_minival',)
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level

--- a/configs/test_time_aug/keypoint_rcnn_R-50-FPN_1x.yaml
+++ b/configs/test_time_aug/keypoint_rcnn_R-50-FPN_1x.yaml
@@ -44,7 +44,7 @@ TEST:
  DATASETS: ('keypoints_coco_2014_minival',)
  PROPOSAL_FILES: ('https://s3-us-west-2.amazonaws.com/detectron/35998996/12_2017_baselines/rpn_person_only_R-50-FPN_1x.yaml.08_10_08.0ZWmJm6F/output/test/keypoints_coco_2014_minival/generalized_rcnn/rpn_proposals.pkl',)
  PROPOSAL_LIMIT: 1000
-  SCALES: (800,)
+  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  WEIGHTS: https://s3-us-west-2.amazonaws.com/detectron/37651887/12_2017_baselines/keypoint_rcnn_R-50-FPN_s1x.yaml.20_01_40.FDjUQ7VX/output/train/keypoints_coco_2014_train%3Akeypoints_coco_2014_valminusminival/generalized_rcnn/model_final.pkl

--- a/lib/core/config.py
+++ b/lib/core/config.py
@@ -213,11 +213,8 @@ __C.TEST.WEIGHTS = b''
 # If multiple datasets are listed, testing is performed on each one sequentially
 __C.TEST.DATASETS = ()

-# Scales to use during testing
-# Each scale is the pixel size of an image's shortest side
-# If multiple scales are given, then all scales are used as in multiscale
-# inference
-__C.TEST.SCALES = (600, )
+# Scale to use during testing
+__C.TEST.SCALE = 600

 # Max pixel size of the longest side of a scaled input image
 __C.TEST.MAX_SIZE = 1000
@@ -968,7 +965,7 @@ __C.CLUSTER.ON_CLUSTER = False
 # yaml configs, you can add the full config key as a string to the set below.
 # ---------------------------------------------------------------------------- #
 _DEPCRECATED_KEYS = set(
-    (
+    {
        'FINAL_MSG',
        'MODEL.DILATION',
        'ROOT_GPU_ID',
@@ -977,7 +974,7 @@ _DEPCRECATED_KEYS = set(
        'TRAIN.DROPOUT',
        'USE_GPU_NMS',
        'TEST.NUM_TEST_IMAGES',
-    )
+    }
 )

 # ---------------------------------------------------------------------------- #
@@ -1006,6 +1003,11 @@ _RENAMED_KEYS = {
        "'path/to/file1:path/to/file2' -> " +
        "('path/to/file1', 'path/to/file2')"
    ),
+    'TEST.SCALES': (
+        'TEST.SCALE',
+        "Also convert from a tuple, e.g. (600, ), " +
+        "to a integer, e.g. 600."
+    ),
 }



--- a/lib/core/rpn_generator.py
+++ b/lib/core/rpn_generator.py
@@ -42,9 +42,9 @@ from core.config import cfg
 from datasets import task_evaluation
 from datasets.json_dataset import JsonDataset
 from modeling import model_builder
-from utils.blob import im_list_to_blob
 from utils.io import save_object
 from utils.timer import Timer
+import utils.blob as blob_utils
 import utils.c2 as c2_utils
 import utils.env as envu
 import utils.net as nu
@@ -186,8 +186,8 @@ def generate_proposals_on_roidb(
 def im_proposals(model, im):
    """Generate RPN proposals on a single image."""
    inputs = {}
-    inputs['data'], inputs['im_info'] = _get_image_blob(im)
-    scale = inputs['im_info'][0, 2]
+    inputs['data'], im_scale, inputs['im_info'] = \
+        blob_utils.get_image_blob_for_inference(im)
    for k, v in inputs.items():
        workspace.FeedBlob(core.ScopedName(k), v.astype(np.float32, copy=False))
    workspace.RunNet(model.net.Proto().name)
@@ -224,7 +224,7 @@ def im_proposals(model, im):
    # Column 0 is the batch index in the (batch ind, x1, y1, x2, y2) encoding,
    # so we remove it since we just want to return boxes
    # Scale proposals back to the original input image scale
-    boxes = boxes[:, 1:] / scale
+    boxes = boxes[:, 1:] / im_scale
    return boxes, scores


@@ -255,41 +255,3 @@ def evaluate_proposal_file(dataset, proposal_file, output_dir):
    recall_file = os.path.join(output_dir, 'rpn_proposal_recall.pkl')
    save_object(results, recall_file)
    return results
-
-
-def _get_image_blob(im):
-    """Converts an image into a network input.
-
-    Arguments:
-        im (ndarray): a color image in BGR order
-
-    Returns:
-        blob (ndarray): a data blob holding an image pyramid
-        im_scale_factors (list): list of image scales (relative to im) used
-            in the image pyramid
-    """
-    im_orig = im.astype(np.float32, copy=True)
-    im_orig -= cfg.PIXEL_MEANS
-
-    im_shape = im_orig.shape
-    im_size_min = np.min(im_shape[0:2])
-    im_size_max = np.max(im_shape[0:2])
-
-    processed_ims = []
-
-    assert len(cfg.TEST.SCALES) == 1
-    target_size = cfg.TEST.SCALES[0]
-
-    im_scale = float(target_size) / float(im_size_min)
-    # Prevent the biggest axis from being more than MAX_SIZE
-    if np.round(im_scale * im_size_max) > cfg.TEST.MAX_SIZE:
-        im_scale = float(cfg.TEST.MAX_SIZE) / float(im_size_max)
-    im = cv2.resize(im_orig, None, None, fx=im_scale, fy=im_scale,
-                    interpolation=cv2.INTER_LINEAR)
-    im_info = np.hstack((im.shape[:2], im_scale))[np.newaxis, :]
-    processed_ims.append(im)
-
-    # Create a blob to hold the input images
-    blob = im_list_to_blob(processed_ims)
-
-    return blob, im_info
--- a/lib/core/test.py
+++ b/lib/core/test.py
@@ -60,9 +60,9 @@ def im_detect_all(model, im, box_proposals, timers=None):

    timers['im_detect_bbox'].tic()
    if cfg.TEST.BBOX_AUG.ENABLED:
-        scores, boxes, im_scales = im_detect_bbox_aug(model, im, box_proposals)
+        scores, boxes, im_scale = im_detect_bbox_aug(model, im, box_proposals)
    else:
-        scores, boxes, im_scales = im_detect_bbox(model, im, box_proposals)
+        scores, boxes, im_scale = im_detect_bbox(model, im, box_proposals)
    timers['im_detect_bbox'].toc()

    # score and boxes are from the whole image after score thresholding and nms
@@ -78,7 +78,7 @@ def im_detect_all(model, im, box_proposals, timers=None):
        if cfg.TEST.MASK_AUG.ENABLED:
            masks = im_detect_mask_aug(model, im, boxes)
        else:
-            masks = im_detect_mask(model, im_scales, boxes)
+            masks = im_detect_mask(model, im_scale, boxes)
        timers['im_detect_mask'].toc()

        timers['misc_mask'].tic()
@@ -94,7 +94,7 @@ def im_detect_all(model, im, box_proposals, timers=None):
        if cfg.TEST.KPS_AUG.ENABLED:
            heatmaps = im_detect_keypoints_aug(model, im, boxes)
        else:
-            heatmaps = im_detect_keypoints(model, im_scales, boxes)
+            heatmaps = im_detect_keypoints(model, im_scale, boxes)
        timers['im_detect_keypoints'].toc()

        timers['misc_keypoints'].tic()
@@ -108,10 +108,10 @@ def im_detect_all(model, im, box_proposals, timers=None):

 def im_conv_body_only(model, im):
    """Runs `model.conv_body_net` on the given image `im`."""
-    im_blob, im_scale_factors = _get_image_blob(im)
+    im_blob, im_scale, _im_info = blob_utils.get_image_blob_for_inference(im)
    workspace.FeedBlob(core.ScopedName('data'), im_blob)
    workspace.RunNet(model.conv_body_net.Proto().name)
-    return im_scale_factors
+    return im_scale


 def im_detect_bbox(model, im, boxes=None):
@@ -130,7 +130,7 @@ def im_detect_bbox(model, im, boxes=None):
        im_scales (list): list of image scales used in the input blob (as
            returned by _get_blobs and for use with im_detect_mask, etc.)
    """
-    inputs, im_scales = _get_blobs(im, boxes)
+    inputs, im_scale = _get_blobs(im, boxes)

    # When mapping from image ROIs to feature map ROIs, there's some aliasing
    # (some distinct image ROIs get mapped to the same feature ROI).
@@ -155,11 +155,9 @@ def im_detect_bbox(model, im, boxes=None):

    # Read out blobs
    if cfg.MODEL.FASTER_RCNN:
-        assert len(im_scales) == 1, \
-            'Only single-image / single-scale batch implemented'
        rois = workspace.FetchBlob(core.ScopedName('rois'))
        # unscale back to raw image space
-        boxes = rois[:, 1:5] / im_scales[0]
+        boxes = rois[:, 1:5] / im_scale

    # Softmax class probabilities
    scores = workspace.FetchBlob(core.ScopedName('cls_prob')).squeeze()
@@ -189,7 +187,7 @@ def im_detect_bbox(model, im, boxes=None):
        scores = scores[inv_index, :]
        pred_boxes = pred_boxes[inv_index, :]

-    return scores, pred_boxes, im_scales
+    return scores, pred_boxes, im_scale


 def im_detect_bbox_aug(model, im, box_proposals=None):
@@ -218,7 +216,7 @@ def im_detect_bbox_aug(model, im, box_proposals=None):

    # Perform detection on the horizontally flipped image
    if cfg.TEST.BBOX_AUG.H_FLIP:
-        scores_hf, boxes_hf, _im_scales_hf = im_detect_bbox_hflip(
+        scores_hf, boxes_hf, _ = im_detect_bbox_hflip(
            model, im, box_proposals
        )
        add_preds_t(scores_hf, boxes_hf)
@@ -253,7 +251,7 @@ def im_detect_bbox_aug(model, im, box_proposals=None):
    # Compute detections for the original image (identity transform) last to
    # ensure that the Caffe2 workspace is populated with blobs corresponding
    # to the original image on return (postcondition of im_detect_bbox)
-    scores_i, boxes_i, im_scales_i = im_detect_bbox(model, im, box_proposals)
+    scores_i, boxes_i, im_scale_i = im_detect_bbox(model, im, box_proposals)
    add_preds_t(scores_i, boxes_i)

    # Combine the predicted scores
@@ -280,7 +278,7 @@ def im_detect_bbox_aug(model, im, box_proposals=None):
            'Coord heur {} not supported'.format(cfg.TEST.BBOX_AUG.COORD_HEUR)
        )

-    return scores_c, boxes_c, im_scales_i
+    return scores_c, boxes_c, im_scale_i


 def im_detect_bbox_hflip(model, im, box_proposals=None):
@@ -296,14 +294,14 @@ def im_detect_bbox_hflip(model, im, box_proposals=None):
    else:
        box_proposals_hf = None

-    scores_hf, boxes_hf, im_scales = im_detect_bbox(
+    scores_hf, boxes_hf, im_scale = im_detect_bbox(
        model, im_hf, box_proposals_hf
    )

    # Invert the detections computed on the flipped image
    boxes_inv = box_utils.flip_boxes(boxes_hf, im_width)

-    return scores_hf, boxes_inv, im_scales
+    return scores_hf, boxes_inv, im_scale


 def im_detect_bbox_scale(
@@ -313,11 +311,11 @@ def im_detect_bbox_scale(
    Returns predictions in the original image space.
    """
    # Remember the original scale
-    orig_scales = cfg.TEST.SCALES
+    orig_scale = cfg.TEST.SCALE
    orig_max_size = cfg.TEST.MAX_SIZE

    # Perform detection at the given scale
-    cfg.TEST.SCALES = (scale, )
+    cfg.TEST.SCALE = scale
    cfg.TEST.MAX_SIZE = max_size

    if hflip:
@@ -328,7 +326,7 @@ def im_detect_bbox_scale(
        scores_scl, boxes_scl, _ = im_detect_bbox(model, im, box_proposals)

    # Restore the original scale
-    cfg.TEST.SCALES = orig_scales
+    cfg.TEST.SCALE = orig_scale
    cfg.TEST.MAX_SIZE = orig_max_size

    return scores_scl, boxes_scl
@@ -361,7 +359,7 @@ def im_detect_bbox_aspect_ratio(
    return scores_ar, boxes_inv


-def im_detect_mask(model, im_scales, boxes):
+def im_detect_mask(model, im_scale, boxes):
    """Infer instance segmentation masks. This function must be called after
    im_detect_bbox as it assumes that the Caffe2 workspace is already populated
    with the necessary blobs.
@@ -377,15 +375,12 @@ def im_detect_mask(model, im_scales, boxes):
            output by the network (must be processed by segm_results to convert
            into hard masks in the original image coordinate space)
    """
-    assert len(im_scales) == 1, \
-        'Only single-image / single-scale batch implemented'
-
    M = cfg.MRCNN.RESOLUTION
    if boxes.shape[0] == 0:
        pred_masks = np.zeros((0, M, M), np.float32)
        return pred_masks

-    inputs = {'mask_rois': _get_rois_blob(boxes, im_scales)}
+    inputs = {'mask_rois': _get_rois_blob(boxes, im_scale)}
    # Add multi-level rois for FPN
    if cfg.FPN.MULTILEVEL_ROIS:
        _add_multilevel_rois_for_test(inputs, 'mask_rois')
@@ -425,8 +420,8 @@ def im_detect_mask_aug(model, im, boxes):
    masks_ts = []

    # Compute masks for the original image (identity transform)
-    im_scales_i = im_conv_body_only(model, im)
-    masks_i = im_detect_mask(model, im_scales_i, boxes)
+    im_scale_i = im_conv_body_only(model, im)
+    masks_i = im_detect_mask(model, im_scale_i, boxes)
    masks_ts.append(masks_i)

    # Perform mask detection on the horizontally flipped image
@@ -486,8 +481,8 @@ def im_detect_mask_hflip(model, im, boxes):
    im_hf = im[:, ::-1, :]
    boxes_hf = box_utils.flip_boxes(boxes, im.shape[1])

-    im_scales = im_conv_body_only(model, im_hf)
-    masks_hf = im_detect_mask(model, im_scales, boxes_hf)
+    im_scale = im_conv_body_only(model, im_hf)
+    masks_hf = im_detect_mask(model, im_scale, boxes_hf)

    # Invert the predicted soft masks
    masks_inv = masks_hf[:, :, :, ::-1]
@@ -499,21 +494,21 @@ def im_detect_mask_scale(model, im, scale, max_size, boxes, hflip=False):
    """Computes masks at the given scale."""

    # Remember the original scale
-    orig_scales = cfg.TEST.SCALES
+    orig_scale = cfg.TEST.SCALE
    orig_max_size = cfg.TEST.MAX_SIZE

    # Perform mask detection at the given scale
-    cfg.TEST.SCALES = (scale, )
+    cfg.TEST.SCALE = scale
    cfg.TEST.MAX_SIZE = max_size

    if hflip:
        masks_scl = im_detect_mask_hflip(model, im, boxes)
    else:
-        im_scales = im_conv_body_only(model, im)
-        masks_scl = im_detect_mask(model, im_scales, boxes)
+        im_scale = im_conv_body_only(model, im)
+        masks_scl = im_detect_mask(model, im_scale, boxes)

    # Restore the original scale
-    cfg.TEST.SCALES = orig_scales
+    cfg.TEST.SCALE = orig_scale
    cfg.TEST.MAX_SIZE = orig_max_size

    return masks_scl
@@ -529,13 +524,13 @@ def im_detect_mask_aspect_ratio(model, im, aspect_ratio, boxes, hflip=False):
    if hflip:
        masks_ar = im_detect_mask_hflip(model, im_ar, boxes_ar)
    else:
-        im_scales = im_conv_body_only(model, im_ar)
-        masks_ar = im_detect_mask(model, im_scales, boxes_ar)
+        im_scale = im_conv_body_only(model, im_ar)
+        masks_ar = im_detect_mask(model, im_scale, boxes_ar)

    return masks_ar


-def im_detect_keypoints(model, im_scales, boxes):
+def im_detect_keypoints(model, im_scale, boxes):
    """Infer instance keypoint poses. This function must be called after
    im_detect_bbox as it assumes that the Caffe2 workspace is already populated
    with the necessary blobs.
@@ -552,15 +547,12 @@ def im_detect_keypoints(model, im_scales, boxes):
            by the network (must be processed by keypoint_results to convert
            into point predictions in the original image coordinate space)
    """
-    assert len(im_scales) == 1, \
-        'Only single-image / single-scale batch implemented'
-
    M = cfg.KRCNN.HEATMAP_SIZE
    if boxes.shape[0] == 0:
        pred_heatmaps = np.zeros((0, cfg.KRCNN.NUM_KEYPOINTS, M, M), np.float32)
        return pred_heatmaps

-    inputs = {'keypoint_rois': _get_rois_blob(boxes, im_scales)}
+    inputs = {'keypoint_rois': _get_rois_blob(boxes, im_scale)}

    # Add multi-level rois for FPN
    if cfg.FPN.MULTILEVEL_ROIS:
@@ -603,8 +595,8 @@ def im_detect_keypoints_aug(model, im, boxes):
        us_ts.append(us_t)

    # Compute the heatmaps for the original image (identity transform)
-    im_scales = im_conv_body_only(model, im)
-    heatmaps_i = im_detect_keypoints(model, im_scales, boxes)
+    im_scale = im_conv_body_only(model, im)
+    heatmaps_i = im_detect_keypoints(model, im_scale, boxes)
    add_heatmaps_t(heatmaps_i)

    # Perform keypoints detection on the horizontally flipped image
@@ -614,8 +606,8 @@ def im_detect_keypoints_aug(model, im, boxes):

    # Compute detections at different scales
    for scale in cfg.TEST.KPS_AUG.SCALES:
-        ds_scl = scale < cfg.TEST.SCALES[0]
-        us_scl = scale > cfg.TEST.SCALES[0]
+        ds_scl = scale < cfg.TEST.SCALE
+        us_scl = scale > cfg.TEST.SCALE
        heatmaps_scl = im_detect_keypoints_scale(
            model, im, scale, cfg.TEST.KPS_AUG.MAX_SIZE, boxes
        )
@@ -672,8 +664,8 @@ def im_detect_keypoints_hflip(model, im, boxes):
    im_hf = im[:, ::-1, :]
    boxes_hf = box_utils.flip_boxes(boxes, im.shape[1])

-    im_scales = im_conv_body_only(model, im_hf)
-    heatmaps_hf = im_detect_keypoints(model, im_scales, boxes_hf)
+    im_scale = im_conv_body_only(model, im_hf)
+    heatmaps_hf = im_detect_keypoints(model, im_scale, boxes_hf)

    # Invert the predicted keypoints
    heatmaps_inv = keypoint_utils.flip_heatmaps(heatmaps_hf)
@@ -685,21 +677,21 @@ def im_detect_keypoints_scale(model, im, scale, max_size, boxes, hflip=False):
    """Computes keypoint predictions at the given scale."""

    # Store the original scale
-    orig_scales = cfg.TEST.SCALES
+    orig_scale = cfg.TEST.SCALE
    orig_max_size = cfg.TEST.MAX_SIZE

    # Perform detection at the given scale
-    cfg.TEST.SCALES = (scale, )
+    cfg.TEST.SCALE = scale
    cfg.TEST.MAX_SIZE = max_size

    if hflip:
        heatmaps_scl = im_detect_keypoints_hflip(model, im, boxes)
    else:
-        im_scales = im_conv_body_only(model, im)
-        heatmaps_scl = im_detect_keypoints(model, im_scales, boxes)
+        im_scale = im_conv_body_only(model, im)
+        heatmaps_scl = im_detect_keypoints(model, im_scale, boxes)

    # Restore the original scale
-    cfg.TEST.SCALES = orig_scales
+    cfg.TEST.SCALE = orig_scale
    cfg.TEST.MAX_SIZE = orig_max_size

    return heatmaps_scl
@@ -717,8 +709,8 @@ def im_detect_keypoints_aspect_ratio(
    if hflip:
        heatmaps_ar = im_detect_keypoints_hflip(model, im_ar, boxes_ar)
    else:
-        im_scales = im_conv_body_only(model, im_ar)
-        heatmaps_ar = im_detect_keypoints(model, im_scales, boxes_ar)
+        im_scale = im_conv_body_only(model, im_ar)
+        heatmaps_ar = im_detect_keypoints(model, im_scale, boxes_ar)

    return heatmaps_ar

@@ -891,25 +883,7 @@ def keypoint_results(cls_boxes, pred_heatmaps, ref_boxes):
    return cls_keyps


-def _get_image_blob(im):
-    """Converts an image into a network input.
-
-    Arguments:
-        im (ndarray): a color image in BGR order
-
-    Returns:
-        blob (ndarray): a data blob holding an image pyramid
-        im_scale_factors (ndarray): array of image scales (relative to im) used
-            in the image pyramid
-    """
-    processed_ims, im_scale_factors = blob_utils.prep_im_for_blob(
-        im, cfg.PIXEL_MEANS, cfg.TEST.SCALES, cfg.TEST.MAX_SIZE
-    )
-    blob = blob_utils.im_list_to_blob(processed_ims)
-    return blob, np.array(im_scale_factors)
-
-
-def _get_rois_blob(im_rois, im_scale_factors):
+def _get_rois_blob(im_rois, im_scale):
    """Converts RoIs into network inputs.

    Arguments:
@@ -920,7 +894,7 @@ def _get_rois_blob(im_rois, im_scale_factors):
        blob (ndarray): R x 5 matrix of RoIs in the image pyramid with columns
            [level, x1, y1, x2, y2]
    """
-    rois, levels = _project_im_rois(im_rois, im_scale_factors)
+    rois, levels = _project_im_rois(im_rois, im_scale)
    rois_blob = np.hstack((levels, rois))
    return rois_blob.astype(np.float32, copy=False)

@@ -936,21 +910,8 @@ def _project_im_rois(im_rois, scales):
        rois (ndarray): R x 4 matrix of projected RoI coordinates
        levels (ndarray): image pyramid levels used by each projected RoI
    """
-    im_rois = im_rois.astype(np.float, copy=False)
-
-    if len(scales) > 1:
-        widths = im_rois[:, 2] - im_rois[:, 0] + 1
-        heights = im_rois[:, 3] - im_rois[:, 1] + 1
-
-        areas = widths * heights
-        scaled_areas = areas[:, np.newaxis] * (scales[np.newaxis, :]**2)
-        diff_areas = np.abs(scaled_areas - 224 * 224)
-        levels = diff_areas.argmin(axis=1)[:, np.newaxis]
-    else:
+    rois = im_rois.astype(np.float, copy=False) * scales
    levels = np.zeros((im_rois.shape[0], 1), dtype=np.int)
-
-    rois = im_rois * scales[levels]
-
    return rois, levels


@@ -978,11 +939,8 @@ def _add_multilevel_rois_for_test(blobs, name):
 def _get_blobs(im, rois):
    """Convert an image and RoIs within that image into network inputs."""
    blobs = {}
-    blobs['data'], im_scale_factors = _get_image_blob(im)
-    if cfg.MODEL.FASTER_RCNN and rois is None:
-        height, width = blobs['data'].shape[2], blobs['data'].shape[3]
-        scale = im_scale_factors[0]
-        blobs['im_info'] = np.array([[height, width, scale]], dtype=np.float32)
+    blobs['data'], im_scale, blobs['im_info'] = \
+        blob_utils.get_image_blob_for_inference(im)
    if rois is not None:
-        blobs['rois'] = _get_rois_blob(rois, im_scale_factors)
-    return blobs, im_scale_factors
+        blobs['rois'] = _get_rois_blob(rois, im_scale)
+    return blobs, im_scale
--- a/lib/core/test_retinanet.py
+++ b/lib/core/test_retinanet.py
@@ -27,10 +27,10 @@ from collections import defaultdict
 from caffe2.python import core, workspace

 from core.config import cfg
-from core.rpn_generator import _get_image_blob
 from modeling.generate_anchors import generate_anchors
 from utils.timer import Timer

+import utils.blob as blob_utils
 import utils.boxes as box_utils

 logger = logging.getLogger(__name__)
@@ -76,7 +76,8 @@ def im_detect_bbox(model, im, timers=None):
    k_max, k_min = cfg.FPN.RPN_MAX_LEVEL, cfg.FPN.RPN_MIN_LEVEL
    A = cfg.RETINANET.SCALES_PER_OCTAVE * len(cfg.RETINANET.ASPECT_RATIOS)
    inputs = {}
-    inputs['data'], inputs['im_info'] = _get_image_blob(im)
+    inputs['data'], im_scale, inputs['im_info'] = \
+        blob_utils.get_image_blob_for_inference(im)
    cls_probs, box_preds = [], []
    for lvl in range(k_min, k_max + 1):
        suffix = 'fpn{}'.format(lvl)
@@ -86,7 +87,6 @@ def im_detect_bbox(model, im, timers=None):
        workspace.FeedBlob(core.ScopedName(k), v.astype(np.float32, copy=False))

    workspace.RunNet(model.net.Proto().name)
-    scale = inputs['im_info'][0, 2]
    cls_probs = workspace.FetchBlobs(cls_probs)
    box_preds = workspace.FetchBlobs(box_preds)

@@ -147,7 +147,7 @@ def im_detect_bbox(model, im, timers=None):
        pred_boxes = (
            box_utils.bbox_transform(boxes, box_deltas)
            if cfg.TEST.BBOX_REG else boxes)
-        pred_boxes /= scale
+        pred_boxes /= im_scale
        pred_boxes = box_utils.clip_tiled_boxes(pred_boxes, im.shape)
        box_scores = np.zeros((pred_boxes.shape[0], 5))
        box_scores[:, 0:4] = pred_boxes

--- a/lib/roi_data/minibatch.py
+++ b/lib/roi_data/minibatch.py
@@ -105,7 +105,7 @@ def _get_image_blob(roidb):
            im = im[:, ::-1, :]
        target_size = cfg.TRAIN.SCALES[scale_inds[i]]
        im, im_scale = blob_utils.prep_im_for_blob(
-            im, cfg.PIXEL_MEANS, [target_size], cfg.TRAIN.MAX_SIZE
+            im, cfg.PIXEL_MEANS, target_size, cfg.TRAIN.MAX_SIZE
        )
        im_scales.append(im_scale[0])
        processed_ims.append(im[0])

--- a/lib/utils/blob.py
+++ b/lib/utils/blob.py
@@ -37,6 +37,33 @@ from caffe2.proto import caffe2_pb2
 from core.config import cfg


+def get_image_blob_for_inference(im):
+    """Converts an image into a network input.
+
+    Arguments:
+        im (ndarray): a color image in BGR order
+
+    Returns:
+        blob (ndarray): a data blob holding an image pyramid
+        im_scale (float): image scale (target size) / (original size)
+        im_info (ndarray)
+    """
+    processed_im, im_scale = prep_im_for_blob(
+        im, cfg.PIXEL_MEANS, cfg.TEST.SCALE, cfg.TEST.MAX_SIZE
+    )
+    blob = im_list_to_blob(processed_im)
+    # NOTE: this height and width may be larger than actual scaled input image
+    # due to the FPN.COARSEST_STRIDE related padding in im_list_to_blob. We are
+    # maintaining this behavior for now to make existing results exactly
+    # reproducible (in practice using the true input image height and width
+    # yields nearly the same results, but they are sometimes slightly different
+    # because predictions near the edge of the image will be pruned more
+    # aggressively).
+    height, width = blob.shape[2], blob.shape[3]
+    im_info = np.hstack((height, width, im_scale))[np.newaxis, :]
+    return blob, im_scale, im_info.astype(np.float32)
+
+
 def im_list_to_blob(ims):
    """Convert a list of images into a network input. Assumes images were
    prepared using prep_im_for_blob or equivalent: i.e.
@@ -47,6 +74,8 @@ def im_list_to_blob(ims):
    Output is a 4D HCHW tensor of the images concatenated along axis 0 with
    shape.
    """
+    if not isinstance(ims, list):
+        ims = [ims]
    max_shape = np.array([im.shape for im in ims]).max(axis=0)
    # Pad the image so they can be divisible by a stride
    if cfg.FPN.FPN_ON:
@@ -55,8 +84,9 @@ def im_list_to_blob(ims):
        max_shape[1] = int(np.ceil(max_shape[1] / stride) * stride)

    num_images = len(ims)
-    blob = np.zeros((num_images, max_shape[0], max_shape[1], 3),
-                    dtype=np.float32)
+    blob = np.zeros(
+        (num_images, max_shape[0], max_shape[1], 3), dtype=np.float32
+    )
    for i in range(num_images):
        im = ims[i]
        blob[i, 0:im.shape[0], 0:im.shape[1], :] = im
@@ -67,7 +97,7 @@ def im_list_to_blob(ims):
    return blob


-def prep_im_for_blob(im, pixel_means, target_sizes, max_size):
+def prep_im_for_blob(im, pixel_means, target_size, max_size):
    """Prepare an image for use as a network input blob. Specially:
      - Subtract per-channel pixel mean
      - Convert to float32
@@ -80,19 +110,19 @@ def prep_im_for_blob(im, pixel_means, target_sizes, max_size):
    im_shape = im.shape
    im_size_min = np.min(im_shape[0:2])
    im_size_max = np.max(im_shape[0:2])
-
-    ims = []
-    im_scales = []
-    for target_size in target_sizes:
    im_scale = float(target_size) / float(im_size_min)
    # Prevent the biggest axis from being more than max_size
    if np.round(im_scale * im_size_max) > max_size:
        im_scale = float(max_size) / float(im_size_max)
-        im = cv2.resize(im, None, None, fx=im_scale, fy=im_scale,
-                        interpolation=cv2.INTER_LINEAR)
-        ims.append(im)
-        im_scales.append(im_scale)
-    return ims, im_scales
+    im = cv2.resize(
+        im,
+        None,
+        None,
+        fx=im_scale,
+        fy=im_scale,
+        interpolation=cv2.INTER_LINEAR
+    )
+    return im, im_scale


 def zeros(shape, int32=False):

--- a/tools/convert_pkl_to_pb.py
+++ b/tools/convert_pkl_to_pb.py
@@ -434,8 +434,6 @@ def _prepare_blobs(


 def run_model_pb(args, net, init_net, im, check_blobs):
-    assert len(cfg.TEST.SCALES) == 1
-
    workspace.ResetWorkspace()
    workspace.RunNetOnce(init_net)
    mutils.create_input_blobs_for_net(net.Proto())
@@ -445,7 +443,7 @@ def run_model_pb(args, net, init_net, im, check_blobs):
    input_blobs = _prepare_blobs(
        im,
        cfg.PIXEL_MEANS,
-        cfg.TEST.SCALES[0], cfg.TEST.MAX_SIZE
+        cfg.TEST.SCALE, cfg.TEST.MAX_SIZE
    )
    gpu_blobs = []
    if args.device == 'gpu':