Merge pull request #311 from wuyefeilin/develop

add resnet_vd

configs/deeplabv3p_resnet50_vd_cityscapes.yaml

+EVAL_CROP_SIZE: (2049, 1025) # (width, height), for unpadding rangescaling and stepscaling
+TRAIN_CROP_SIZE: (769, 769) # (width, height), for unpadding rangescaling and stepscaling
+AUG:
+    AUG_METHOD: "stepscaling" # choice unpadding rangescaling and stepscaling
+    FIX_RESIZE_SIZE: (2048, 1024) # (width, height), for unpadding
+    INF_RESIZE_VALUE: 500  # for rangescaling
+    MAX_RESIZE_VALUE: 600  # for rangescaling
+    MIN_RESIZE_VALUE: 400  # for rangescaling
+    MAX_SCALE_FACTOR: 2.0  # for stepscaling
+    MIN_SCALE_FACTOR: 0.5  # for stepscaling
+    SCALE_STEP_SIZE: 0.25  # for stepscaling
+    MIRROR: True
+    TO_RGB: True
+BATCH_SIZE: 16
+DATASET:
+    DATA_DIR: "./dataset/cityscapes/"
+    IMAGE_TYPE: "rgb"  # choice rgb or rgba
+    NUM_CLASSES: 19
+    TEST_FILE_LIST: "dataset/cityscapes/val.list"
+    TRAIN_FILE_LIST: "dataset/cityscapes/train.list"
+    VAL_FILE_LIST: "dataset/cityscapes/val.list"
+    IGNORE_INDEX: 255
+    SEPARATOR: " "
+FREEZE:
+    MODEL_FILENAME: "model"
+    PARAMS_FILENAME: "params"
+MODEL:
+    DEFAULT_NORM_TYPE: "bn"
+    MODEL_NAME: "deeplabv3p"
+    DEEPLAB:
+        ASPP_WITH_SEP_CONV: True
+        DECODER_USE_SEP_CONV: True
+        BACKBONE: "resnet_vd_50"
+        BACKBONE_LR_MULT_LIST: [0.1, 0.1, 0.2, 0.2, 1.0]
+TRAIN:
+    PRETRAINED_MODEL_DIR: u"pretrained_model/resnet50_vd_imagenet"
+    MODEL_SAVE_DIR: "saved_model/deeplabv3p_resnet50_vd_bn_cityscapes"
+    SNAPSHOT_EPOCH: 10
+    SYNC_BATCH_NORM: True
+TEST:
+    TEST_MODEL: "saved_model/deeplabv3p_resnet50_vd_bn_cityscapes/final"
+SOLVER:
+    LR: 0.05
+    LR_POLICY: "poly"
+    OPTIMIZER: "sgd"
+    NUM_EPOCHS: 700

docs/model_zoo.md

@@ -37,6 +37,7 @@ PaddleSeg对所有内置的分割模型都提供了公开数据集下的预训
 |---|---|---|---|
 | ResNet50（适配PSPNet） | ImageNet | [resnet50_v2_pspnet](https://paddleseg.bj.bcebos.com/resnet50_v2_pspnet.tgz)| -- |
 | ResNet101（适配PSPNet） | ImageNet | [resnet101_v2_pspnet](https://paddleseg.bj.bcebos.com/resnet101_v2_pspnet.tgz)| -- |
+| ResNet50_vd | ImageNet | [ResNet50_vd_ssld_pretrained.tgz](https://paddleseg.bj.bcebos.com/models/ResNet50_vd_ssld_pretrained.tgz) | 83.0%/96.4% |
 
 ## COCO预训练模型
 
@@ -58,7 +59,8 @@ train数据集合为Cityscapes训练集合，测试为Cityscapes的验证集合
 |---|---|---|---|---|---|
 | DeepLabv3+/MobileNetv2/bn | Cityscapes |[mobilenet_cityscapes.tgz](https://paddleseg.bj.bcebos.com/models/mobilenet_cityscapes.tgz) |16|false| 0.698|
 | DeepLabv3+/Xception65/gn  | Cityscapes |[deeplabv3p_xception65_gn_cityscapes.tgz](https://paddleseg.bj.bcebos.com/models/deeplabv3p_xception65_cityscapes.tgz) |16|false| 0.7824 |
-| DeepLabv3+/Xception65/bn | Cityscapes |[deeplabv3p_xception65_bn_cityscapes_.tgz](https://paddleseg.bj.bcebos.com/models/xception65_bn_cityscapes.tgz) | 16 | false | 0.7930 |
+| DeepLabv3+/Xception65/bn | Cityscapes |[deeplabv3p_xception65_bn_cityscapes.tgz](https://paddleseg.bj.bcebos.com/models/xception65_bn_cityscapes.tgz) | 16 | false | 0.7930 |
+| DeepLabv3+/ResNet50_vd/bn | Cityscapes |[deeplabv3p_resnet50_vd_cityscapes.tgz](https://paddleseg.bj.bcebos.com/models/deeplabv3p_resnet50_vd_cityscapes.tgz) | 16 | false | 0.8006 |
 | ICNet/bn | Cityscapes |[icnet_cityscapes.tgz](https://paddleseg.bj.bcebos.com/models/icnet_cityscapes.tar.gz) |16|false| 0.6831 |
 | PSPNet/bn | Cityscapes |[pspnet50_cityscapes.tgz](https://paddleseg.bj.bcebos.com/models/pspnet50_cityscapes.tgz) |16|false| 0.7013 |
 | PSPNet/bn | Cityscapes |[pspnet101_cityscapes.tgz](https://paddleseg.bj.bcebos.com/models/pspnet101_cityscapes.tgz) |16|false| 0.7734 |

pdseg/models/modeling/deeplab.py

@@ -352,6 +352,8 @@ def resnet_vd(input):
     else:
         raise Exception("deeplab only support stride 8 or 16")
     lr_mult_list = cfg.MODEL.DEEPLAB.BACKBONE_LR_MULT_LIST
+    if lr_mult_list is None:
+        lr_mult_list = [1.0, 1.0, 1.0, 1.0, 1.0]
     model = resnet_vd_backbone(
         layers, stem='deeplab', lr_mult_list=lr_mult_list)
     data, decode_shortcuts = model.net(

pretrained_model/download_model.py

@@ -42,6 +42,8 @@ model_urls = {
     "https://paddleseg.bj.bcebos.com/models/Xception41_pretrained.tgz",
     "xception65_imagenet":
     "https://paddleseg.bj.bcebos.com/models/Xception65_pretrained.tgz",
+    "resnet50_vd_imagenet":
+    "https://paddleseg.bj.bcebos.com/models/ResNet50_vd_ssld_pretrained.tgz",
     "hrnet_w18_bn_imagenet":
     "https://paddleseg.bj.bcebos.com/models/hrnet_w18_imagenet.tar",
     "hrnet_w30_bn_imagenet":
@@ -76,6 +78,8 @@ model_urls = {
     "https://paddleseg.bj.bcebos.com/models/deeplabv3p_xception65_cityscapes.tgz",
     "deeplabv3p_xception65_bn_cityscapes":
     "https://paddleseg.bj.bcebos.com/models/xception65_bn_cityscapes.tgz",
+    "deeplabv3p_resnet50_vd_cityscapes":
+    "https://paddleseg.bj.bcebos.com/models/deeplabv3p_resnet50_vd_cityscapes.tgz",
     "unet_bn_coco":
     "https://paddleseg.bj.bcebos.com/models/unet_coco_v3.tgz",
     "icnet_bn_cityscapes":

turtorial/finetune_deeplabv3plus.md

@@ -145,8 +145,10 @@ PaddleSeg在AI Studio平台上提供了在线体验的DeepLabv3+图像分割教
 |mobilenetv2-0-25_bn_imagenet|MobileNetV2|bn|ImageNet|MODEL.MODEL_NAME: deeplabv3p <br> MODEL.DEEPLAB.BACKBONE: mobilenetv2 <br> MODEL.DEEPLAB.DEPTH_MULTIPLIER: 0.25 <br> MODEL.DEFAULT_NORM_TYPE: bn|
 |xception41_imagenet|Xception41|bn|ImageNet|MODEL.MODEL_NAME: deeplabv3p <br> MODEL.DEEPLAB.BACKBONE: xception_41 <br> MODEL.DEFAULT_NORM_TYPE: bn|
 |xception65_imagenet|Xception65|bn|ImageNet|MODEL.MODEL_NAME: deeplabv3p <br> MODEL.DEEPLAB.BACKBONE: xception_65 <br> MODEL.DEFAULT_NORM_TYPE: bn|
+|resnet50_vd_imagenet|ResNet50_vd|bn|ImageNet|MODEL.MODEL_NAME: deeplabv3p <br> MODEL.DEEPLAB.BACKBONE: resnet50_vd <br> MODEL.DEFAULT_NORM_TYPE: bn|
 |deeplabv3p_mobilenetv2-1-0_bn_coco|MobileNetV2|bn|COCO|MODEL.MODEL_NAME: deeplabv3p <br> MODEL.DEEPLAB.BACKBONE: mobilenetv2 <br> MODEL.DEEPLAB.DEPTH_MULTIPLIER: 1.0 <br> MODEL.DEEPLAB.ENCODER_WITH_ASPP: False <br> MODEL.DEEPLAB.ENABLE_DECODER: False <br> MODEL.DEFAULT_NORM_TYPE: bn|
 |**deeplabv3p_xception65_bn_coco**|Xception65|bn|COCO|MODEL.MODEL_NAME: deeplabv3p <br> MODEL.DEEPLAB.BACKBONE: xception_65 <br> MODEL.DEFAULT_NORM_TYPE: bn |
 |deeplabv3p_mobilenetv2-1-0_bn_cityscapes|MobileNetV2|bn|Cityscapes|MODEL.MODEL_NAME: deeplabv3p <br> MODEL.DEEPLAB.BACKBONE: mobilenetv2 <br> MODEL.DEEPLAB.DEPTH_MULTIPLIER: 1.0 <br> MODEL.DEEPLAB.ENCODER_WITH_ASPP: False <br> MODEL.DEEPLAB.ENABLE_DECODER: False <br> MODEL.DEFAULT_NORM_TYPE: bn|
 |deeplabv3p_xception65_gn_cityscapes|Xception65|gn|Cityscapes|MODEL.MODEL_NAME: deeplabv3p <br>  MODEL.DEEPLAB.BACKBONE: xception_65 <br> MODEL.DEFAULT_NORM_TYPE: gn|
 |deeplabv3p_xception65_bn_cityscapes|Xception65|bn|Cityscapes|MODEL.MODEL_NAME: deeplabv3p <br> MODEL.DEEPLAB.BACKBONE: xception_65 <br> MODEL.DEFAULT_NORM_TYPE: bn|
+|deeplabv3p_resnet50_vd_cityscapes|resnet50_vd|bn|Cityscapes|MODEL.MODEL_NAME: deeplabv3p <br> MODEL.DEEPLAB.BACKBONE: resnet50_vd <br> MODEL.DEFAULT_NORM_TYPE: bn|