Merge pull request facebookresearch#2 from chengyangfu/retina_maskrcnn

Retina maskrcnn

Author: Cheng-Yang Fu

ABSTRACTIONS.md

@@ -31,15 +31,15 @@ a specific image, as well as the size of the image as a `(width, height)` tuple.
 It also contains a set of methods that allow to perform geometric
 transformations to the bounding boxes (such as cropping, scaling and flipping).
 The class accepts bounding boxes from two different input formats:
-- `xyxy`, where each box is encoded as a `x1`, `y1`, `x2` and `y2` coordinates)
+- `xyxy`, where each box is encoded as a `x1`, `y1`, `x2` and `y2` coordinates, and
 - `xywh`, where each box is encoded as `x1`, `y1`, `w` and `h`.
 
 Additionally, each `BoxList` instance can also hold arbitrary additional information
 for each bounding box, such as labels, visibility, probability scores etc.
 
 Here is an example on how to create a `BoxList` from a list of coordinates:
 ```python
-from maskrcnn_baseline.structures.bounding_box import BoxList, FLIP_LEFT_RIGHT
+from maskrcnn_benchmark.structures.bounding_box import BoxList, FLIP_LEFT_RIGHT
 
 width = 100
 height = 200
@@ -49,7 +49,7 @@ boxes = [
   [10, 10, 50, 50]
 ]
 # create a BoxList with 3 boxes
-bbox = BoxList(boxes, size=(width, height), mode='xyxy')
+bbox = BoxList(boxes, image_size=(width, height), mode='xyxy')
 
 # perform some box transformations, has similar API as PIL.Image
 bbox_scaled = bbox.resize((width * 2, height * 3))

configs/retina/retinanet_R-101-FPN_1x.yaml

@@ -0,0 +1,48 @@
+MODEL:
+  META_ARCHITECTURE: "RetinaNet"
+  WEIGHT: "catalog://ImageNetPretrained/MSRA/R-101"
+  RPN_ONLY: True
+  BACKBONE:
+    CONV_BODY: "R-101-FPN"
+    OUT_CHANNELS: 256
+  RPN:
+    USE_FPN: True
+    FG_IOU_THRESHOLD: 0.5
+    BG_IOU_THRESHOLD: 0.4
+    ANCHOR_STRIDE: (4, 8, 16, 32, 64)
+    PRE_NMS_TOP_N_TRAIN: 2000
+    PRE_NMS_TOP_N_TEST: 1000
+    POST_NMS_TOP_N_TEST: 1000
+    FPN_POST_NMS_TOP_N_TEST: 1000
+  ROI_HEADS:
+    USE_FPN: True
+    BATCH_SIZE_PER_IMAGE: 256
+  ROI_BOX_HEAD:
+    POOLER_RESOLUTION: 7
+    POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125)
+    POOLER_SAMPLING_RATIO: 2
+    FEATURE_EXTRACTOR: "FPN2MLPFeatureExtractor"
+    PREDICTOR: "FPNPredictor"
+DATASETS:
+  TRAIN: ("coco_2017_train",)
+  TEST: ("coco_2017_val",)
+INPUT:
+  MIN_SIZE_TRAIN: (800, )
+  MAX_SIZE_TRAIN: 1333
+  MIN_SIZE_TEST: 800
+  MAX_SIZE_TEST: 1333
+DATALOADER:
+  SIZE_DIVISIBILITY: 32
+SOLVER:
+  # Assume 4 gpus
+  BASE_LR: 0.005
+  WEIGHT_DECAY: 0.0001
+  STEPS: (120000, 160000)
+  MAX_ITER: 180000
+  IMS_PER_BATCH: 8
+RETINANET:
+  RETINANET_ON: True
+  SCALES_PER_OCTAVE: 3
+  STRADDLE_THRESH: -1
+
+

configs/retina/retinanet_R-50-FPN_1x.yaml

@@ -0,0 +1,46 @@
+MODEL:
+  META_ARCHITECTURE: "RetinaNet"
+  WEIGHT: "catalog://ImageNetPretrained/MSRA/R-50"
+  RPN_ONLY: True
+  BACKBONE:
+    CONV_BODY: "R-50-FPN"
+    OUT_CHANNELS: 256
+  RPN:
+    USE_FPN: True
+    FG_IOU_THRESHOLD: 0.5
+    BG_IOU_THRESHOLD: 0.4
+    ANCHOR_STRIDE: (4, 8, 16, 32, 64)
+    PRE_NMS_TOP_N_TRAIN: 2000
+    PRE_NMS_TOP_N_TEST: 1000
+    POST_NMS_TOP_N_TEST: 1000
+    FPN_POST_NMS_TOP_N_TEST: 1000
+  ROI_HEADS:
+    USE_FPN: True
+    BATCH_SIZE_PER_IMAGE: 256
+  ROI_BOX_HEAD:
+    POOLER_RESOLUTION: 7
+    POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125)
+    POOLER_SAMPLING_RATIO: 2
+    FEATURE_EXTRACTOR: "FPN2MLPFeatureExtractor"
+    PREDICTOR: "FPNPredictor"
+DATASETS:
+  TRAIN: ("coco_2017_train",)
+  TEST: ("coco_2017_val",)
+INPUT:
+  MIN_SIZE_TRAIN: (800,)
+  MAX_SIZE_TRAIN: 1333
+  MIN_SIZE_TEST: 800
+  MAX_SIZE_TEST: 1333
+DATALOADER:
+  SIZE_DIVISIBILITY: 32
+SOLVER:
+  # Assume 4 gpus
+  BASE_LR: 0.01
+  WEIGHT_DECAY: 0.0001
+  STEPS: (60000, 80000)
+  MAX_ITER: 90000
+  IMS_PER_BATCH: 16
+RETINANET:
+  RETINANET_ON: True
+  SCALES_PER_OCTAVE: 3
+  STRADDLE_THRESH: -1

configs/retina/retinanet_R-50-FPN_1x_adjust_std011.yaml

@@ -0,0 +1,47 @@
+MODEL:
+  META_ARCHITECTURE: "RetinaNet"
+  WEIGHT: "catalog://ImageNetPretrained/MSRA/R-50"
+  RPN_ONLY: True
+  BACKBONE:
+    CONV_BODY: "R-50-FPN"
+    OUT_CHANNELS: 256
+  RPN:
+    USE_FPN: True
+    FG_IOU_THRESHOLD: 0.5
+    BG_IOU_THRESHOLD: 0.4
+    ANCHOR_STRIDE: (4, 8, 16, 32, 64)
+    PRE_NMS_TOP_N_TRAIN: 2000
+    PRE_NMS_TOP_N_TEST: 1000
+    POST_NMS_TOP_N_TEST: 1000
+    FPN_POST_NMS_TOP_N_TEST: 1000
+  ROI_HEADS:
+    USE_FPN: True
+    BATCH_SIZE_PER_IMAGE: 256
+  ROI_BOX_HEAD:
+    POOLER_RESOLUTION: 7
+    POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125)
+    POOLER_SAMPLING_RATIO: 2
+    FEATURE_EXTRACTOR: "FPN2MLPFeatureExtractor"
+    PREDICTOR: "FPNPredictor"
+DATASETS:
+  TRAIN: ("coco_2017_train",)
+  TEST: ("coco_2017_val",)
+INPUT:
+  MIN_SIZE_TRAIN: (800,)
+  MAX_SIZE_TRAIN: 1333
+  MIN_SIZE_TEST: 800
+  MAX_SIZE_TEST: 1333
+DATALOADER:
+  SIZE_DIVISIBILITY: 32
+SOLVER:
+  # Assume 4 gpus
+  BASE_LR: 0.01
+  WEIGHT_DECAY: 0.0001
+  STEPS: (60000, 80000)
+  MAX_ITER: 90000
+  IMS_PER_BATCH: 16
+RETINANET:
+  RETINANET_ON: True
+  SCALES_PER_OCTAVE: 3
+  STRADDLE_THRESH: -1
+  SELFADJUST_SMOOTH_L1: True

configs/retina/retinanet_R-50-FPN_1x_adjust_std100.yaml

@@ -0,0 +1,48 @@
+MODEL:
+  META_ARCHITECTURE: "RetinaNet"
+  WEIGHT: "catalog://ImageNetPretrained/MSRA/R-50"
+  RPN_ONLY: True
+  BACKBONE:
+    CONV_BODY: "R-50-FPN"
+    OUT_CHANNELS: 256
+  RPN:
+    USE_FPN: True
+    FG_IOU_THRESHOLD: 0.5
+    BG_IOU_THRESHOLD: 0.4
+    ANCHOR_STRIDE: (4, 8, 16, 32, 64)
+    PRE_NMS_TOP_N_TRAIN: 2000
+    PRE_NMS_TOP_N_TEST: 1000
+    POST_NMS_TOP_N_TEST: 1000
+    FPN_POST_NMS_TOP_N_TEST: 1000
+  ROI_HEADS:
+    USE_FPN: True
+    BATCH_SIZE_PER_IMAGE: 256
+  ROI_BOX_HEAD:
+    POOLER_RESOLUTION: 7
+    POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125)
+    POOLER_SAMPLING_RATIO: 2
+    FEATURE_EXTRACTOR: "FPN2MLPFeatureExtractor"
+    PREDICTOR: "FPNPredictor"
+DATASETS:
+  TRAIN: ("coco_2017_train",)
+  TEST: ("coco_2017_val",)
+INPUT:
+  MIN_SIZE_TRAIN: (800,)
+  MAX_SIZE_TRAIN: 1333
+  MIN_SIZE_TEST: 800
+  MAX_SIZE_TEST: 1333
+DATALOADER:
+  SIZE_DIVISIBILITY: 32
+SOLVER:
+  # Assume 4 gpus
+  BASE_LR: 0.01
+  WEIGHT_DECAY: 0.0001
+  STEPS: (60000, 80000)
+  MAX_ITER: 90000
+  IMS_PER_BATCH: 16
+RETINANET:
+  RETINANET_ON: True
+  SCALES_PER_OCTAVE: 3
+  STRADDLE_THRESH: -1
+  BBOX_REG_BETA: 1.0
+  SELFADJUST_SMOOTH_L1: True

configs/retina/retinanet_R-50-FPN_1x_adjustl1.yaml

@@ -0,0 +1,47 @@
+MODEL:
+  META_ARCHITECTURE: "RetinaNet"
+  WEIGHT: "catalog://ImageNetPretrained/MSRA/R-50"
+  RPN_ONLY: True
+  BACKBONE:
+    CONV_BODY: "R-50-FPN"
+    OUT_CHANNELS: 256
+  RPN:
+    USE_FPN: True
+    FG_IOU_THRESHOLD: 0.5
+    BG_IOU_THRESHOLD: 0.4
+    ANCHOR_STRIDE: (4, 8, 16, 32, 64)
+    PRE_NMS_TOP_N_TRAIN: 2000
+    PRE_NMS_TOP_N_TEST: 1000
+    POST_NMS_TOP_N_TEST: 1000
+    FPN_POST_NMS_TOP_N_TEST: 1000
+  ROI_HEADS:
+    USE_FPN: True
+    BATCH_SIZE_PER_IMAGE: 256
+  ROI_BOX_HEAD:
+    POOLER_RESOLUTION: 7
+    POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125)
+    POOLER_SAMPLING_RATIO: 2
+    FEATURE_EXTRACTOR: "FPN2MLPFeatureExtractor"
+    PREDICTOR: "FPNPredictor"
+DATASETS:
+  TRAIN: ("coco_2017_train",)
+  TEST: ("coco_2017_val",)
+INPUT:
+  MIN_SIZE_TRAIN: (800,)
+  MAX_SIZE_TRAIN: 1333
+  MIN_SIZE_TEST: 800
+  MAX_SIZE_TEST: 1333
+DATALOADER:
+  SIZE_DIVISIBILITY: 32
+SOLVER:
+  # Assume 4 gpus
+  BASE_LR: 0.01
+  WEIGHT_DECAY: 0.0001
+  STEPS: (60000, 80000)
+  MAX_ITER: 90000
+  IMS_PER_BATCH: 16
+RETINANET:
+  RETINANET_ON: True
+  SCALES_PER_OCTAVE: 3
+  STRADDLE_THRESH: -1
+  SELFADJUST_SMOOTH_L1: True

configs/retina/retinanet_R-50-FPN_1x_beta100.yaml

@@ -0,0 +1,48 @@
+MODEL:
+  META_ARCHITECTURE: "RetinaNet"
+  WEIGHT: "catalog://ImageNetPretrained/MSRA/R-50"
+  RPN_ONLY: True
+  BACKBONE:
+    CONV_BODY: "R-50-FPN"
+    OUT_CHANNELS: 256
+  RPN:
+    USE_FPN: True
+    FG_IOU_THRESHOLD: 0.5
+    BG_IOU_THRESHOLD: 0.4
+    ANCHOR_STRIDE: (4, 8, 16, 32, 64)
+    PRE_NMS_TOP_N_TRAIN: 2000
+    PRE_NMS_TOP_N_TEST: 1000
+    POST_NMS_TOP_N_TEST: 1000
+    FPN_POST_NMS_TOP_N_TEST: 1000
+  ROI_HEADS:
+    USE_FPN: True
+    BATCH_SIZE_PER_IMAGE: 256
+  ROI_BOX_HEAD:
+    POOLER_RESOLUTION: 7
+    POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125)
+    POOLER_SAMPLING_RATIO: 2
+    FEATURE_EXTRACTOR: "FPN2MLPFeatureExtractor"
+    PREDICTOR: "FPNPredictor"
+DATASETS:
+  TRAIN: ("coco_2017_train",)
+  TEST: ("coco_2017_val",)
+INPUT:
+  MIN_SIZE_TRAIN: (800,)
+  MAX_SIZE_TRAIN: 1333
+  MIN_SIZE_TEST: 800
+  MAX_SIZE_TEST: 1333
+DATALOADER:
+  SIZE_DIVISIBILITY: 32
+SOLVER:
+  # Assume 4 gpus
+  BASE_LR: 0.01
+  WEIGHT_DECAY: 0.0001
+  STEPS: (60000, 80000)
+  MAX_ITER: 90000
+  IMS_PER_BATCH: 16
+RETINANET:
+  RETINANET_ON: True
+  SCALES_PER_OCTAVE: 3
+  STRADDLE_THRESH: -1
+  BBOX_REG_BETA: 1.0
+  SELFADJUST_SMOOTH_L1: False

configs/retina/retinanet_R-50-FPN_1x_low_quality_0.2.yaml

@@ -0,0 +1,47 @@
+MODEL:
+  META_ARCHITECTURE: "RetinaNet"
+  WEIGHT: "catalog://ImageNetPretrained/MSRA/R-50"
+  RPN_ONLY: True
+  BACKBONE:
+    CONV_BODY: "R-50-FPN"
+    OUT_CHANNELS: 256
+  RPN:
+    USE_FPN: True
+    FG_IOU_THRESHOLD: 0.5
+    BG_IOU_THRESHOLD: 0.4
+    ANCHOR_STRIDE: (4, 8, 16, 32, 64)
+    PRE_NMS_TOP_N_TRAIN: 2000
+    PRE_NMS_TOP_N_TEST: 1000
+    POST_NMS_TOP_N_TEST: 1000
+    FPN_POST_NMS_TOP_N_TEST: 1000
+  ROI_HEADS:
+    USE_FPN: True
+    BATCH_SIZE_PER_IMAGE: 256
+  ROI_BOX_HEAD:
+    POOLER_RESOLUTION: 7
+    POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125)
+    POOLER_SAMPLING_RATIO: 2
+    FEATURE_EXTRACTOR: "FPN2MLPFeatureExtractor"
+    PREDICTOR: "FPNPredictor"
+DATASETS:
+  TRAIN: ("coco_2017_train",)
+  TEST: ("coco_2017_val",)
+INPUT:
+  MIN_SIZE_TRAIN: (800,)
+  MAX_SIZE_TRAIN: 1333
+  MIN_SIZE_TEST: 800
+  MAX_SIZE_TEST: 1333
+DATALOADER:
+  SIZE_DIVISIBILITY: 32
+SOLVER:
+  # Assume 4 gpus
+  BASE_LR: 0.01
+  WEIGHT_DECAY: 0.0001
+  STEPS: (60000, 80000)
+  MAX_ITER: 90000
+  IMS_PER_BATCH: 16
+RETINANET:
+  RETINANET_ON: True
+  SCALES_PER_OCTAVE: 3
+  STRADDLE_THRESH: -1
+  LOW_QUALITY_THRESHOLD: 0.4