PASCAL VOC 2012 support & improvements

configs/coco.yaml

@@ -0,0 +1,58 @@
+EXP:
+    ID: coco
+    OUTPUT_DIR: data
+
+DATASET:
+    NAME: coco
+    ROOT:
+    LABELS: ./data/datasets/coco/labels.txt
+    N_CLASSES: 91
+    IGNORE_LABEL:
+    SCALES:
+    SPLIT:
+        TRAIN:
+        VAL:
+        TEST:
+
+DATALOADER:
+    NUM_WORKERS: 0
+
+IMAGE:
+    MEAN:
+        R: 122.675
+        G: 116.669
+        B: 104.008
+    SIZE:
+        BASE:
+        TRAIN:
+        TEST: 513
+
+MODEL:
+    NAME: DeepLabV1_ResNet101
+    N_BLOCKS: [3, 4, 23, 3]
+    ATROUS_RATES:
+    INIT_MODEL:
+
+SOLVER:
+    BATCH_SIZE:
+        TRAIN: 5
+        TEST: 1
+    ITER_MAX: 100000
+    ITER_SIZE: 2
+    ITER_SAVE: 5000
+    ITER_TB: 100
+    LR_DECAY: 10
+    LR: 2.5e-4
+    MOMENTUM: 0.9
+    OPTIMIZER: sgd
+    POLY_POWER: 0.9
+    WEIGHT_DECAY: 5.0e-4
+    AVERAGE_LOSS: 20
+
+CRF:
+    ITER_MAX: 10
+    POS_W: 3
+    POS_XY_STD: 1
+    BI_W: 4
+    BI_XY_STD: 67
+    BI_RGB_STD: 3

config/cocostuff10k.yaml → configs/cocostuff10k.yaml

@@ -1,10 +1,13 @@
+EXP:
+    ID: cocostuff10k
+    OUTPUT_DIR: data
+
 DATASET:
     NAME: cocostuff10k
     ROOT: /media/kazuto1011/Extra/cocostuff/cocostuff-10k-v1.1
-    LABELS: ./data/datasets/cocostuff/labels_2.txt
-    N_CLASSES: 182    
-    IGNORE_LABEL: -1
-    WARP_IMAGE: True
+    LABELS: ./data/datasets/cocostuff/labels.txt
+    N_CLASSES: 182
+    IGNORE_LABEL: 255
     SCALES: [0.5, 0.75, 1.0, 1.25, 1.5]
     SPLIT:
         TRAIN: train
@@ -20,17 +23,15 @@ IMAGE:
         G: 116.669
         B: 104.008
     SIZE:
-        TRAIN:
-            BASE: 513
-            CROP: 321
+        BASE:
+        TRAIN: 321
         TEST: 513
 
 MODEL:
     NAME: DeepLabV2_ResNet101_MSC
     N_BLOCKS: [3, 4, 23, 3]
     ATROUS_RATES: [6, 12, 18, 24]
-    INIT_MODEL: ./data/models/deeplab_resnet101/coco_init/deeplabv2_resnet101_COCO_init.pth
-    SAVE_DIR: ./data/models/deeplab_resnet101/cocostuff10k
+    INIT_MODEL: data/models/coco/deeplabv1_resnet101/caffemodel/deeplabv1_resnet101-coco.pth
 
 SOLVER:
     BATCH_SIZE:
@@ -46,7 +47,6 @@ SOLVER:
     OPTIMIZER: sgd
     POLY_POWER: 0.9
     WEIGHT_DECAY: 5.0e-4
-    LOG_DIR: runs/cocostuff10k
     AVERAGE_LOSS: 20
 
 CRF:
@@ -55,4 +55,4 @@ CRF:
     POS_XY_STD: 1
     BI_W: 4
     BI_XY_STD: 67
-    BI_RGB_STD: 3
+    BI_RGB_STD: 3

config/cocostuff164k.yaml → configs/cocostuff164k.yaml

@@ -1,10 +1,13 @@
+EXP:
+    ID: cocostuff164k
+    OUTPUT_DIR: data
+
 DATASET:
     NAME: cocostuff164k
     ROOT: /media/kazuto1011/Extra/cocostuff/cocostuff-164k
-    LABELS: ./data/datasets/cocostuff/labels_2.txt
+    LABELS: ./data/datasets/cocostuff/labels.txt
     N_CLASSES: 182
     IGNORE_LABEL: 255
-    WARP_IMAGE: True
     SCALES: [0.5, 0.75, 1.0, 1.25, 1.5]
     SPLIT:
         TRAIN: train2017
@@ -20,22 +23,20 @@ IMAGE:
         G: 116.669
         B: 104.008
     SIZE:
-        TRAIN:
-            BASE: 513
-            CROP: 321
+        BASE: # None
+        TRAIN: 321
         TEST: 513
 
 MODEL:
     NAME: DeepLabV2_ResNet101_MSC
     N_BLOCKS: [3, 4, 23, 3]
     ATROUS_RATES: [6, 12, 18, 24]
-    INIT_MODEL: ./data/models/deeplab_resnet101/coco_init/deeplabv2_resnet101_COCO_init.pth
-    SAVE_DIR: ./data/models/deeplab_resnet101/cocostuff164k
+    INIT_MODEL: data/models/coco/deeplabv1_resnet101/caffemodel/deeplabv1_resnet101-coco.pth
 
 SOLVER:
     BATCH_SIZE:
         TRAIN: 5
-        TEST: 5
+        TEST: 1
     ITER_MAX: 100000
     ITER_SIZE: 2
     ITER_SAVE: 5000
@@ -46,7 +47,6 @@ SOLVER:
     OPTIMIZER: sgd
     POLY_POWER: 0.9
     WEIGHT_DECAY: 5.0e-4
-    LOG_DIR: runs/cocostuff164k
     AVERAGE_LOSS: 20
 
 CRF:
@@ -55,4 +55,4 @@ CRF:
     POS_XY_STD: 1
     BI_W: 4
     BI_XY_STD: 67
-    BI_RGB_STD: 3
+    BI_RGB_STD: 3

configs/conda_env.yaml

@@ -0,0 +1,26 @@
+name: deeplab-pytorch
+channels:
+  - pytorch
+  - conda-forge
+  - defaults
+dependencies:
+  # - clang # For MacOS
+  # - clangxx # For MacOS
+  - click
+  - cudatoolkit=10.0
+  - cython
+  - matplotlib
+  - pytorch
+  - pyyaml
+  - scipy
+  - torchvision
+  - tqdm
+  - pip:
+    - addict
+    - black
+    - joblib
+    - opencv-python
+    - pydensecrf
+    - tensorboardX
+    - tensorflow
+    - torchnet

config/voc12.yaml → configs/voc12.yaml

@@ -1,15 +1,18 @@
+EXP:
+    ID: voc12
+    OUTPUT_DIR: data
+
 DATASET:
-    NAME: voc12
-    ROOT:
+    NAME: vocaug
+    ROOT: /media/kazuto1011/Extra/VOCdevkit
     LABELS: ./data/datasets/voc12/labels.txt
     N_CLASSES: 21
     IGNORE_LABEL: 255
-    WARP_IMAGE: False
     SCALES: [0.5, 0.75, 1.0, 1.25, 1.5]
     SPLIT:
-        TRAIN:
-        VAL:
-        TEST:
+        TRAIN: train_aug
+        VAL: val
+        TEST: test
 
 DATALOADER:
     NUM_WORKERS: 0
@@ -20,22 +23,20 @@ IMAGE:
         G: 116.669
         B: 104.008
     SIZE:
-        TRAIN:
-            BASE: 513
-            CROP: 321
+        BASE: # None
+        TRAIN: 321
         TEST: 513
 
 MODEL:
     NAME: DeepLabV2_ResNet101_MSC
     N_BLOCKS: [3, 4, 23, 3]
     ATROUS_RATES: [6, 12, 18, 24]
-    INIT_MODEL: ./data/models/deeplab_resnet101/coco_init/deeplabv2_resnet101_COCO_init.pth
-    SAVE_DIR: ./data/models/deeplab_resnet101/voc12
+    INIT_MODEL: data/models/coco/deeplabv1_resnet101/caffemodel/deeplabv1_resnet101-coco.pth
 
 SOLVER:
     BATCH_SIZE:
         TRAIN: 5
-        TEST: 5
+        TEST: 1
     ITER_MAX: 20000
     ITER_SIZE: 2
     ITER_SAVE: 5000
@@ -46,7 +47,6 @@ SOLVER:
     OPTIMIZER: sgd
     POLY_POWER: 0.9
     WEIGHT_DECAY: 5.0e-4
-    LOG_DIR: runs/voc12
     AVERAGE_LOSS: 20
 
 CRF:
@@ -55,4 +55,4 @@ CRF:
     POS_XY_STD: 1
     BI_W: 4
     BI_XY_STD: 67
-    BI_RGB_STD: 3
+    BI_RGB_STD: 3