Add at new repo again

vton-api/preprocess/humanparsing/mhp_extension/detectron2/projects/PointRend/configs/InstanceSegmentation/Base-PointRend-RCNN-FPN.yaml

@@ -0,0 +1,21 @@
+_BASE_: "../../../../configs/Base-RCNN-FPN.yaml"
+MODEL:
+  ROI_HEADS:
+    NAME: "PointRendROIHeads"
+    IN_FEATURES: ["p2", "p3", "p4", "p5"]
+  ROI_BOX_HEAD:
+    TRAIN_ON_PRED_BOXES: True
+  ROI_MASK_HEAD:
+    NAME: "CoarseMaskHead"
+    FC_DIM: 1024
+    NUM_FC: 2
+    OUTPUT_SIDE_RESOLUTION: 7
+    IN_FEATURES: ["p2"]
+    POINT_HEAD_ON: True
+  POINT_HEAD:
+    FC_DIM: 256
+    NUM_FC: 3
+    IN_FEATURES: ["p2"]
+INPUT:
+  # PointRend for instance segmenation does not work with "polygon" mask_format.
+  MASK_FORMAT: "bitmask"

vton-api/preprocess/humanparsing/mhp_extension/detectron2/projects/PointRend/configs/InstanceSegmentation/pointrend_rcnn_R_50_FPN_1x_cityscapes.yaml

@@ -0,0 +1,23 @@
+_BASE_: Base-PointRend-RCNN-FPN.yaml
+MODEL:
+  WEIGHTS: detectron2://ImageNetPretrained/MSRA/R-50.pkl
+  MASK_ON: true
+  RESNETS:
+    DEPTH: 50
+  ROI_HEADS:
+    NUM_CLASSES: 8
+  POINT_HEAD:
+    NUM_CLASSES: 8
+DATASETS:
+  TEST: ("cityscapes_fine_instance_seg_val",)
+  TRAIN: ("cityscapes_fine_instance_seg_train",)
+SOLVER:
+  BASE_LR: 0.01
+  IMS_PER_BATCH: 8
+  MAX_ITER: 24000
+  STEPS: (18000,)
+INPUT:
+  MAX_SIZE_TEST: 2048
+  MAX_SIZE_TRAIN: 2048
+  MIN_SIZE_TEST: 1024
+  MIN_SIZE_TRAIN: (800, 832, 864, 896, 928, 960, 992, 1024)

vton-api/preprocess/humanparsing/mhp_extension/detectron2/projects/PointRend/configs/InstanceSegmentation/pointrend_rcnn_R_50_FPN_1x_coco.yaml

@@ -0,0 +1,9 @@
+_BASE_: Base-PointRend-RCNN-FPN.yaml
+MODEL:
+  WEIGHTS: detectron2://ImageNetPretrained/MSRA/R-50.pkl
+  MASK_ON: true
+  RESNETS:
+    DEPTH: 50
+# To add COCO AP evaluation against the higher-quality LVIS annotations.
+# DATASETS:
+#   TEST: ("coco_2017_val", "lvis_v0.5_val_cocofied")

vton-api/preprocess/humanparsing/mhp_extension/detectron2/projects/PointRend/configs/InstanceSegmentation/pointrend_rcnn_R_50_FPN_3x_coco.yaml

@@ -0,0 +1,13 @@
+_BASE_: Base-PointRend-RCNN-FPN.yaml
+MODEL:
+  WEIGHTS: detectron2://ImageNetPretrained/MSRA/R-50.pkl
+  MASK_ON: true
+  RESNETS:
+    DEPTH: 50
+SOLVER:
+  STEPS: (210000, 250000)
+  MAX_ITER: 270000
+# To add COCO AP evaluation against the higher-quality LVIS annotations.
+# DATASETS:
+#   TEST: ("coco_2017_val", "lvis_v0.5_val_cocofied")
+

vton-api/preprocess/humanparsing/mhp_extension/detectron2/projects/PointRend/configs/InstanceSegmentation/pointrend_rcnn_R_50_FPN_3x_parsing.yaml

@@ -0,0 +1,20 @@
+_BASE_: Base-PointRend-RCNN-FPN.yaml
+MODEL:
+  WEIGHTS: detectron2://ImageNetPretrained/MSRA/R-50.pkl
+  MASK_ON: true
+  RESNETS:
+    DEPTH: 50
+  ROI_HEADS:
+    NUM_CLASSES: 1
+  POINT_HEAD:
+    NUM_CLASSES: 1
+SOLVER:
+  STEPS: (210000, 250000)
+  MAX_ITER: 270000
+  IMS_PER_BATCH: 1
+# To add COCO AP evaluation against the higher-quality LVIS annotations.
+# DATASETS:
+#   TEST: ("coco_2017_val", "lvis_v0.5_val_cocofied")
+DATASETS:
+  TRAIN: ("CIHP_train",)
+  TEST: ("CIHP_val",)

vton-api/preprocess/humanparsing/mhp_extension/detectron2/projects/PointRend/configs/InstanceSegmentation/pointrend_rcnn_X_101_32x8d_FPN_3x_parsing.yaml

@@ -0,0 +1,28 @@
+_BASE_: Base-PointRend-RCNN-FPN.yaml
+MODEL:
+  WEIGHTS: "./X-101-32x8d.pkl"
+  PIXEL_STD: [57.375, 57.120, 58.395]
+  MASK_ON: true
+  RESNETS:
+    STRIDE_IN_1X1: False  # this is a C2 model
+    NUM_GROUPS: 32
+    WIDTH_PER_GROUP: 8
+    DEPTH: 101
+  ROI_HEADS:
+    NUM_CLASSES: 1
+  POINT_HEAD:
+    NUM_CLASSES: 1
+SOLVER:
+  STEPS: (210000, 250000)
+  MAX_ITER: 270000
+  IMS_PER_BATCH: 1
+# To add COCO AP evaluation against the higher-quality LVIS annotations.
+# DATASETS:
+#   TEST: ("coco_2017_val", "lvis_v0.5_val_cocofied")
+INPUT:
+  MIN_SIZE_TRAIN: (640, 864)
+  MIN_SIZE_TRAIN_SAMPLING: "range"
+  MAX_SIZE_TRAIN: 1440
+DATASETS:
+  TRAIN: ("CIHP_train",)
+  TEST: ("CIHP_val",)

vton-api/preprocess/humanparsing/mhp_extension/detectron2/projects/PointRend/configs/SemanticSegmentation/Base-PointRend-Semantic-FPN.yaml

@@ -0,0 +1,19 @@
+_BASE_: "../../../../configs/Base-RCNN-FPN.yaml"
+MODEL:
+  META_ARCHITECTURE: "SemanticSegmentor"
+  BACKBONE:
+    FREEZE_AT: 0
+  SEM_SEG_HEAD:
+    NAME: "PointRendSemSegHead"
+  POINT_HEAD:
+    NUM_CLASSES: 54
+    FC_DIM: 256
+    NUM_FC: 3
+    IN_FEATURES: ["p2"]
+    TRAIN_NUM_POINTS: 1024
+    SUBDIVISION_STEPS: 2
+    SUBDIVISION_NUM_POINTS: 8192
+    COARSE_SEM_SEG_HEAD_NAME: "SemSegFPNHead"
+DATASETS:
+  TRAIN: ("coco_2017_train_panoptic_stuffonly",)
+  TEST: ("coco_2017_val_panoptic_stuffonly",)

vton-api/preprocess/humanparsing/mhp_extension/detectron2/projects/PointRend/configs/SemanticSegmentation/pointrend_semantic_R_101_FPN_1x_cityscapes.yaml

@@ -0,0 +1,33 @@
+_BASE_: Base-PointRend-Semantic-FPN.yaml
+MODEL:
+  WEIGHTS: detectron2://ImageNetPretrained/MSRA/R-101.pkl
+  RESNETS:
+    DEPTH: 101
+  SEM_SEG_HEAD:
+    NUM_CLASSES: 19
+  POINT_HEAD:
+    NUM_CLASSES: 19
+    TRAIN_NUM_POINTS: 2048
+    SUBDIVISION_NUM_POINTS: 8192
+DATASETS:
+  TRAIN: ("cityscapes_fine_sem_seg_train",)
+  TEST: ("cityscapes_fine_sem_seg_val",)
+SOLVER:
+  BASE_LR: 0.01
+  STEPS: (40000, 55000)
+  MAX_ITER: 65000
+  IMS_PER_BATCH: 32
+INPUT:
+  MIN_SIZE_TRAIN: (512, 768, 1024, 1280, 1536, 1792, 2048)
+  MIN_SIZE_TRAIN_SAMPLING: "choice"
+  MIN_SIZE_TEST: 1024
+  MAX_SIZE_TRAIN: 4096
+  MAX_SIZE_TEST: 2048
+  CROP:
+    ENABLED: True
+    TYPE: "absolute"
+    SIZE: (512, 1024)
+    SINGLE_CATEGORY_MAX_AREA: 0.75
+  COLOR_AUG_SSD: True
+DATALOADER:
+  NUM_WORKERS: 16

vton-api/preprocess/humanparsing/mhp_extension/detectron2/projects/PointRend/configs/SemanticSegmentation/pointrend_semantic_R_50_FPN_1x_coco.yaml

@@ -0,0 +1,5 @@
+_BASE_: Base-PointRend-Semantic-FPN.yaml
+MODEL:
+  WEIGHTS: detectron2://ImageNetPretrained/MSRA/R-50.pkl
+  RESNETS:
+    DEPTH: 50