Update

chenjoya · Aug 25, 2021 · 02a3846 · 02a3846
1 parent 1e93d9a
commit 02a3846
Show file tree

Hide file tree

Showing 192 changed files with 2,528 additions and 7,628 deletions.
diff --git a/.flake8 b/.flake8
diff --git a/ABSTRACTIONS.md b/ABSTRACTIONS.md
diff --git a/CODE_OF_CONDUCT.md b/CODE_OF_CONDUCT.md
diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
diff --git a/INSTALL.md b/INSTALL.md
@@ -15,10 +15,7 @@ conda create --name sampling-free python=3.8
 conda activate sampling-free
 
 # sampling-free and cocoapi dependencies
-pip install ninja yacs cython matplotlib tqdm opencv-python pillow
-
-# install cocoapi
-pip install pycocotools
+pip install ninja yacs cython matplotlib tqdm opencv-python pillow pycocotools
 
 # follow PyTorch installation in https://pytorch.org/get-started/locally/
 conda install pytorch torchvision cudatoolkit=11.1

diff --git a/LICENSE b/LICENSE
diff --git a/README.md b/README.md
@@ -56,14 +56,14 @@ author    = {Joya Chen and
              Dong Liu and
              Tong Xu and
              Shiwei Wu and
-             Yifei Chen and
+             Yifei Cheng and
              Enhong Chen},
 title     = {Is Heuristic Sampling Necessary in Training Deep Object Detectors?},
-journal   = {CoRR},
-volume    = {abs/1909.04868},
-url       = {http://arxiv.org/abs/1909.04868},
-archivePrefix = {arXiv},
-eprint    = {1909.04868},
+journal   = {IEEE Transactions on Image Processing},
+year      = {2021},
+volume    = {},
+number    = {},
+pages     = {1-1},
 }
 ```
 

diff --git a/TROUBLESHOOTING.md b/TROUBLESHOOTING.md
diff --git a/configs/ovsampler/faster_rcnn_R_50_FPN_1x.yaml b/configs/ovsampler/faster_rcnn_R_50_FPN_1x.yaml
@@ -0,0 +1,37 @@
+MODEL:
+  ARCHITECTURE: "GeneralizedRCNN"
+  PRETRAINED: "catalog://ImageNetPretrained/MSRA/R-50"
+  RCNN_ON: True
+  OVSAMPLER_ON: True
+  BACKBONE:
+    CONV_BODY: "R-50-FPN"
+  RESNETS:
+    BACKBONE_OUT_CHANNELS: 256
+  RPN:
+    USE_FPN: True
+    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
+    PRIOR_PROB: 1e-4 
+  ROI_HEADS:
+    USE_FPN: True
+    SCORE_THRESH: 1e-3
+  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",)
+DATALOADER:
+  SIZE_DIVISIBILITY: 32
+SOLVER:
+  BATCH_SIZE: 16
+  LR: 0.02
+  WEIGHT_DECAY: 0.0001
+  STEPS: (60000, 80000)
+  MAX_ITER: 90000
diff --git a/configs/ovsampler/mask_rcnn_R_50_FPN_1x.yaml b/configs/ovsampler/mask_rcnn_R_50_FPN_1x.yaml
@@ -0,0 +1,45 @@
+MODEL:
+  ARCHITECTURE: "GeneralizedRCNN"
+  PRETRAINED: "catalog://ImageNetPretrained/MSRA/R-50"
+  RCNN_ON: True
+  OVSAMPLER_ON: True
+  BACKBONE:
+    CONV_BODY: "R-50-FPN"
+  RESNETS:
+    BACKBONE_OUT_CHANNELS: 256
+  RPN:
+    USE_FPN: True
+    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
+    PRIOR_PROB: 1e-4 
+  ROI_HEADS:
+    USE_FPN: True
+    SCORE_THRESH: 1e-3 
+  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"
+  ROI_MASK_HEAD:
+    POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125)
+    FEATURE_EXTRACTOR: "MaskRCNNFPNFeatureExtractor"
+    PREDICTOR: "MaskRCNNC4Predictor"
+    POOLER_RESOLUTION: 14
+    POOLER_SAMPLING_RATIO: 2
+    RESOLUTION: 28
+  MASK_ON: True
+DATASETS:
+  TRAIN: ("coco_2017_train", )
+  TEST: ("coco_2017_val",)
+DATALOADER:
+  SIZE_DIVISIBILITY: 32
+SOLVER:
+  BATCH_SIZE: 16
+  LR: 0.02
+  WEIGHT_DECAY: 0.0001
+  STEPS: (60000, 80000)
+  MAX_ITER: 90000
diff --git a/configs/sampling_free/atss_R_50_FPN_1x.yaml b/configs/sampling_free/atss_R_50_FPN_1x.yaml
@@ -0,0 +1,38 @@
+MODEL:
+  ARCHITECTURE: "GeneralizedRCNN"
+  PRETRAINED: "catalog://ImageNetPretrained/MSRA/R-50"
+  ATSS_ON: True
+  SAMPLING_FREE: True
+  BACKBONE:
+    CONV_BODY: "R-50-FPN-RETINANET"
+  RESNETS:
+    BACKBONE_OUT_CHANNELS: 256
+  RETINANET:
+    USE_C5: False
+  ATSS:
+    ANCHOR_SIZES: (64, 128, 256, 512, 1024) # 8S
+    ASPECT_RATIOS: (1.0,)
+    SCALES_PER_OCTAVE: 1
+    USE_DCN_IN_TOWER: False
+    POSITIVE_TYPE: 'ATSS'
+    TOPK: 9
+    REGRESSION_TYPE: 'BOX'
+    INFERENCE_TH: 1e-3
+    PRIOR_PROB: 3e-3
+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:
+  LR: 0.01
+  WEIGHT_DECAY: 0.0001
+  STEPS: (60000, 80000)
+  MAX_ITER: 90000
+  BATCH_SIZE: 16
+  WARMUP_METHOD: "constant"