add new model

0d97cc8c · Sugon_ldc · 0d97cc8c · 0d97cc8c · 0d97cc8c · 0d97cc8c
Commit 0d97cc8c authored Jun 07, 2023 by Sugon_ldc
20 changed files
--- a/configs/espnetv1/espnetv1_cityscapes_1024x512_120k.yml
+++ b/configs/espnetv1/espnetv1_cityscapes_1024x512_120k.yml
+_base_: '../_base_/cityscapes.yml'
+
+batch_size: 4
+iters: 120000
+
+optimizer:
+  _inherited_: False
+  type: adam
+  weight_decay: 0.0002
+
+lr_scheduler:
+  type: PolynomialDecay
+  learning_rate: 0.001
+  end_lr: 0.0
+  power: 0.9
+
+loss:
+  types:
+    - type: CrossEntropyLoss
+      weight: [2.79834108 ,6.92945723 ,3.84068512 ,9.94349362 ,9.77098823 ,9.51484 ,10.30981624 ,9.94307377 ,4.64933892 ,9.55759938 ,7.86692178 ,9.53126629 ,10.3496365 ,6.67234062 ,10.26054204 ,10.28785275 ,10.28988296 ,10.40546021 ,10.13848367]
+  coef: [1]
+
+model:
+  type: ESPNetV1
+  in_channels: 3
+  num_classes: 19
+  level2_depth: 2
+  level3_depth: 8
--- a/configs/fastfcn/README.md
+++ b/configs/fastfcn/README.md
+# FastFCN: Rethinking Dilated Convolution in the Backbone for Semantic Segmentation
+
+## Reference
+> Wu, Huikai, Junge Zhang, Kaiqi Huang, Kongming Liang, and Yizhou Yu. "Fastfcn: Rethinking dilated convolution in the backbone for semantic segmentation." arXiv preprint arXiv:1903.11816 (2019).
+
+## Performance
+
+### ADE20K
+
+| Model | Backbone | Resolution | Training Iters | mIoU | mIoU (flip) | mIoU (ms+flip) | Links |
+|:-:|:-:|:-:|:-:|:-:|:-:|:-:|:-:|
+|FastFCN|ResNet50_vd|480x480|120000|43.76%|44.11%|44.48%|[model](https://bj.bcebos.com/paddleseg/dygraph/ade20k/fastfcn_resnet50_os8_ade20k_480x480_120k/model.pdparams) \|[log](https://bj.bcebos.com/paddleseg/dygraph/ade20k/fastfcn_resnet50_os8_ade20k_480x480_120k/train.log)\|[vdl](https://www.paddlepaddle.org.cn/paddle/visualdl/service/app/scalar?id=e159d5be3860b8d08762c0416ab54acc)|
--- a/configs/fastfcn/fastfcn_resnet50_os8_ade20k_480x480_120k.yml
+++ b/configs/fastfcn/fastfcn_resnet50_os8_ade20k_480x480_120k.yml
+_base_: '../_base_/ade20k.yml'
+batch_size: 4
+iters: 120000
+
+train_dataset:
+  transforms:
+    - type: ResizeStepScaling
+      min_scale_factor: 0.5
+      max_scale_factor: 2.0
+      scale_step_size: 0.25
+    - type: RandomPaddingCrop
+      crop_size: [480, 480]
+    - type: RandomHorizontalFlip
+    - type: RandomDistort
+      brightness_range: 0.4
+      contrast_range: 0.4
+      saturation_range: 0.4
+    - type: Normalize
+
+model:
+  type: FastFCN
+  backbone:
+    type: ResNet50_vd
+    output_stride: 8
+    pretrained: https://bj.bcebos.com/paddleseg/dygraph/resnet50_vd_ssld_v2.tar.gz
+  num_codes: 32
+  mid_channels: 512
+  use_jpu: True
+  aux_loss: True
+  use_se_loss: True
+  add_lateral: True
+
+loss:
+  types:
+    - type: CrossEntropyLoss
+    - type: CrossEntropyLoss
+    - type: SECrossEntropyLoss
+  coef: [1, 0.4, 0.2]
+
+lr_scheduler:
+  type: PolynomialDecay
+  learning_rate: 0.01
+  end_lr: 0
+  power: 0.9
--- a/configs/fastscnn/README.md
+++ b/configs/fastscnn/README.md
+# Fast-SCNN: Fast Semantic Segmentation Network
+
+## Reference
+
+> Poudel, Rudra PK, Stephan Liwicki, and Roberto Cipolla. "Fast-scnn: Fast semantic segmentation network." arXiv preprint arXiv:1902.04502 (2019).
+
+## Performance
+
+### Cityscapes
+
+| Model | Backbone | Resolution | Training Iters | mIoU | mIoU (flip) | mIoU (ms+flip) | Links |
+|:-:|:-:|:-:|:-:|:-:|:-:|:-:|:-:|
+|Fast SCNN|-|1024x1024|160000|69.31%|-|-|[model](https://bj.bcebos.com/paddleseg/dygraph/cityscapes/fastscnn_cityscapes_1024x1024_160k/model.pdparams) \| [log](https://bj.bcebos.com/paddleseg/dygraph/cityscapes/fastscnn_cityscapes_1024x1024_160k/train.log) \| [vdl](https://paddlepaddle.org.cn/paddle/visualdl/service/app/scalar?id=3b4c3f01c9213cac14e53c69d262a337)|
--- a/configs/fastscnn/fastscnn_cityscapes_1024x1024_160k.yml
+++ b/configs/fastscnn/fastscnn_cityscapes_1024x1024_160k.yml
+_base_: '../_base_/cityscapes_1024x1024.yml'
+
+batch_size: 4
+iters: 160000
+
+loss:
+  types:
+    - type: CrossEntropyLoss
+  coef: [1.0, 0.4]
+
+lr_scheduler:
+  type: PolynomialDecay
+  learning_rate: 0.05
+  end_lr: 1.0e-4
+  power: 0.9
+
+model:
+  type: FastSCNN
+  num_classes: 19
+  enable_auxiliary_loss: True
+  pretrained: null
--- a/configs/fastscnn/fastscnn_cityscapes_1024x1024_40k.yml
+++ b/configs/fastscnn/fastscnn_cityscapes_1024x1024_40k.yml
+_base_: '../_base_/cityscapes_1024x1024.yml'
+
+batch_size: 4
+iters: 40000
+
+loss:
+  types:
+    - type: CrossEntropyLoss
+  coef: [1.0, 0.4]
+
+lr_scheduler:
+  type: PolynomialDecay
+  learning_rate: 0.025
+  end_lr: 1.0e-4
+  power: 0.9
+
+model:
+  type: FastSCNN
+  num_classes: 19
+  enable_auxiliary_loss: True
+  pretrained: null
--- a/configs/fastscnn/fastscnn_cityscapes_1024x1024_40k_SCL.yml
+++ b/configs/fastscnn/fastscnn_cityscapes_1024x1024_40k_SCL.yml
+_base_: '../_base_/cityscapes_1024x1024.yml'
+
+batch_size: 4
+iters: 40000
+
+loss:
+  types:
+    - type: MixedLoss
+      losses:
+        - type: CrossEntropyLoss
+        - type: SemanticConnectivityLoss
+      coef: [1, 0.01]
+    - type: CrossEntropyLoss
+  coef: [1.0, 0.4]
+
+lr_scheduler:
+  type: PolynomialDecay
+  learning_rate: 0.025
+  end_lr: 1.0e-4
+  power: 0.9
+
+model:
+  type: FastSCNN
+  num_classes: 19
+  enable_auxiliary_loss: True
+  pretrained: null
--- a/configs/fcn/README.md
+++ b/configs/fcn/README.md
+# Deep High-Resolution Representation Learning for Visual Recognition
+
+## Reference
+> Wang, Jingdong, Ke Sun, Tianheng Cheng, Borui Jiang, Chaorui Deng, Yang Zhao, Dong Liu et al. "Deep high-resolution representation learning for visual recognition." IEEE transactions on pattern analysis and machine intelligence (2020).
+
+## Performance
+
+### Cityscapes
+
+| Model | Backbone | Resolution | Training Iters | mIoU | mIoU (flip) | mIoU (ms+flip) | Links |
+|:-:|:-:|:-:|:-:|:-:|:-:|:-:|:-:|
+|FCN|HRNet_W18|1024x512|80000|78.97%|79.49%|79.74%|[model](https://bj.bcebos.com/paddleseg/dygraph/cityscapes/fcn_hrnetw18_cityscapes_1024x512_80k/model.pdparams) \| [log](https://bj.bcebos.com/paddleseg/dygraph/cityscapes/fcn_hrnetw18_cityscapes_1024x512_80k/train.log) \| [vdl](https://paddlepaddle.org.cn/paddle/visualdl/service/app?id=bebec8e1a3802c4babd3c69e1bf50d51)|
+|FCN|HRNet_W48|1024x512|80000|80.70%|81.24%|81.56%|[model](https://bj.bcebos.com/paddleseg/dygraph/cityscapes/fcn_hrnetw48_cityscapes_1024x512_80k/model.pdparams) \| [log](https://bj.bcebos.com/paddleseg/dygraph/cityscapes/fcn_hrnetw48_cityscapes_1024x512_80k/train.log) \| [vdl](https://paddlepaddle.org.cn/paddle/visualdl/service/app?id=ae1cb76014cdc54406c36f1e3dc2a530)|
+
+
+### Pascal VOC 2012 + Aug
+
+| Model | Backbone | Resolution | Training Iters | mIoU | mIoU (flip) | mIoU (ms+flip) | Links |
+|:-:|:-:|:-:|:-:|:-:|:-:|:-:|:-:|
+|FCN|HRNet_W18|512x512|40000|75.39%|76.04%|77.09%|[model](https://bj.bcebos.com/paddleseg/dygraph/pascal_voc12/fcn_hrnetw18_voc12aug_512x512_40k/model.pdparams) \| [log](https://bj.bcebos.com/paddleseg/dygraph/pascal_voc12/fcn_hrnetw18_voc12aug_512x512_40k/train.log) \| [vdl](https://paddlepaddle.org.cn/paddle/visualdl/service/app?id=fbe6caaca0f7d7ea1dba1c60b8db2a7e)|
+|FCN|HRNet_W48|512x512|40000|78.72%|79.52%|80.10%|[model](https://bj.bcebos.com/paddleseg/dygraph/pascal_voc12/fcn_hrnetw48_voc12aug_512x512_40k/model.pdparams) \| [log](https://bj.bcebos.com/paddleseg/dygraph/pascal_voc12/fcn_hrnetw48_voc12aug_512x512_40k/train.log) \| [vdl](https://paddlepaddle.org.cn/paddle/visualdl/service/app?id=20b404212fcbb5b7b329ab0c16124553)|
--- a/configs/fcn/fcn_hrnetw18_cityscapes_1024x512_80k.yml
+++ b/configs/fcn/fcn_hrnetw18_cityscapes_1024x512_80k.yml
+_base_: '../_base_/cityscapes.yml'
+
+model:
+  type: FCN
+  backbone:
+    type: HRNet_W18
+    align_corners: False
+    pretrained: https://bj.bcebos.com/paddleseg/dygraph/hrnet_w18_ssld.tar.gz
+  num_classes: 19
+  pretrained: Null
+  backbone_indices: [-1]
+
+optimizer:
+  weight_decay: 0.0005
+
+iters: 80000
--- a/configs/fcn/fcn_hrnetw18_cityscapes_1024x512_80k_bs4.yml
+++ b/configs/fcn/fcn_hrnetw18_cityscapes_1024x512_80k_bs4.yml
+_base_: '../_base_/cityscapes.yml'
+
+model:
+  type: FCN
+  backbone:
+    type: HRNet_W18
+    align_corners: False
+    pretrained: https://bj.bcebos.com/paddleseg/dygraph/hrnet_w18_ssld.tar.gz
+  num_classes: 19
+  pretrained: Null
+  backbone_indices: [-1]
+
+optimizer:
+  weight_decay: 0.0005
+
+iters: 80000
+batch_size: 4
--- a/configs/fcn/fcn_hrnetw18_cityscapes_1024x512_80k_bs4_SCL.yml
+++ b/configs/fcn/fcn_hrnetw18_cityscapes_1024x512_80k_bs4_SCL.yml
+_base_: '../_base_/cityscapes.yml'
+
+model:
+  type: FCN
+  backbone:
+    type: HRNet_W18
+    align_corners: False
+    pretrained: https://bj.bcebos.com/paddleseg/dygraph/hrnet_w18_ssld.tar.gz
+  num_classes: 19
+  pretrained: Null
+  backbone_indices: [-1]
+
+optimizer:
+  weight_decay: 0.0005
+
+iters: 80000
+batch_size: 4
+
+loss:
+  types:
+    - type: MixedLoss
+      losses:
+        - type: CrossEntropyLoss
+        - type: SemanticConnectivityLoss
+      coef: [1, 0.05]
+  coef: [1]
--- a/configs/fcn/fcn_hrnetw18_pphumanseg14k.yml
+++ b/configs/fcn/fcn_hrnetw18_pphumanseg14k.yml
+train_dataset:
+  type: PPHumanSeg14K
+  dataset_root: data/PP-HumanSeg14K
+  transforms:
+    - type: ResizeStepScaling
+      min_scale_factor: 0.5
+      max_scale_factor: 2.0
+      scale_step_size: 0.25
+    - type: RandomPaddingCrop
+      crop_size: [398, 224]
+    - type: RandomHorizontalFlip
+    - type: RandomDistort
+      brightness_range: 0.4
+      contrast_range: 0.4
+      saturation_range: 0.4
+    - type: Normalize
+  mode: train
+
+val_dataset:
+  type: PPHumanSeg14K
+  dataset_root: data/PP-HumanSeg14K
+  transforms:
+    - type: Normalize
+  mode: val
+
+model:
+  type: FCN
+  backbone:
+    type: HRNet_W18
+    align_corners: False
+    pretrained: https://bj.bcebos.com/paddleseg/dygraph/hrnet_w18_ssld.tar.gz
+  num_classes: 2
+  pretrained: Null
+  backbone_indices: [-1]
+
+optimizer:
+  type: sgd
+  momentum: 0.9
+  weight_decay: 0.0005
+
+lr_scheduler:
+  type: PolynomialDecay
+  learning_rate: 0.05
+  end_lr: 0
+  power: 0.9
+
+loss:
+  types:
+    - type: CrossEntropyLoss
+  coef: [1]
+
+iters: 10000
+batch_size: 64
--- a/configs/fcn/fcn_hrnetw18_voc12aug_512x512_40k.yml
+++ b/configs/fcn/fcn_hrnetw18_voc12aug_512x512_40k.yml
+_base_: '../_base_/pascal_voc12aug.yml'
+
+model:
+  type: FCN
+  backbone:
+    type: HRNet_W18
+    align_corners: False
+    pretrained: https://bj.bcebos.com/paddleseg/dygraph/hrnet_w18_ssld.tar.gz
+  num_classes: 21
+  pretrained: Null
+  backbone_indices: [-1]
+
+optimizer:
+  weight_decay: 0.0005
--- a/configs/fcn/fcn_hrnetw48_cityscapes_1024x512_80k.yml
+++ b/configs/fcn/fcn_hrnetw48_cityscapes_1024x512_80k.yml
+_base_: './fcn_hrnetw18_cityscapes_1024x512_80k.yml'
+
+model:
+  backbone:
+    type: HRNet_W48
+    pretrained: https://bj.bcebos.com/paddleseg/dygraph/hrnet_w48_ssld.tar.gz
--- a/configs/fcn/fcn_hrnetw48_voc12aug_512x512_40k.yml
+++ b/configs/fcn/fcn_hrnetw48_voc12aug_512x512_40k.yml
+_base_: './fcn_hrnetw18_voc12aug_512x512_40k.yml'
+
+model:
+  backbone:
+    type: HRNet_W48
+    pretrained: https://bj.bcebos.com/paddleseg/dygraph/hrnet_w48_ssld.tar.gz
--- a/configs/gcnet/README.md
+++ b/configs/gcnet/README.md
+# GCNet: Non-local networks meet squeeze-excitation networks and beyond
+
+## Reference
+
+> Cao, Yue, Jiarui Xu, Stephen Lin, Fangyun Wei, and Han Hu. "GCNet: Non-local networks meet squeeze-excitation networks and beyond." In Proceedings of the IEEE International Conference on Computer Vision Workshops, pp. 0-0. 2019.
+
+## Performance
+
+### Cityscapes
+
+| Model | Backbone | Resolution | Training Iters | mIoU | mIoU (flip) | mIoU (ms+flip) | Links |
+|:-:|:-:|:-:|:-:|:-:|:-:|:-:|:-:|
+|GCNet|ResNet50_OS8|1024x512|80000|79.50%|79.77%|79.69%|[model](https://bj.bcebos.com/paddleseg/dygraph/cityscapes/gcnet_resnet50_os8_cityscapes_1024x512_80k/model.pdparams) \| [log](https://bj.bcebos.com/paddleseg/dygraph/cityscapes/gcnet_resnet50_os8_cityscapes_1024x512_80k/train.log) \| [vdl](https://paddlepaddle.org.cn/paddle/visualdl/service/app?id=e3801edb9a6f5b33eb890f5a1ae6ed7b)|
+|GCNet|ResNet101_OS8|1024x512|80000|81.01%|81.30%|81.64%|[model](https://bj.bcebos.com/paddleseg/dygraph/cityscapes/gcnet_resnet101_os8_cityscapes_1024x512_80k/model.pdparams) \| [log](https://bj.bcebos.com/paddleseg/dygraph/cityscapes/gcnet_resnet101_os8_cityscapes_1024x512_80k/train.log) \| [vdl](https://www.paddlepaddle.org.cn/paddle/visualdl/service/app/scalar?id=aa88e7980f4d6839537662a3a3d18851)|
+
+### Pascal VOC 2012 + Aug
+
+| Model | Backbone | Resolution | Training Iters | mIoU | mIoU (flip) | mIoU (ms+flip) | Links |
+|:-:|:-:|:-:|:-:|:-:|:-:|:-:|:-:|
+|GCNet|ResNet50_OS8|512x512|40000|80.32%|80.39%|80.54%|[model](https://bj.bcebos.com/paddleseg/dygraph/pascal_voc12/gcnet_resnet50_os8_voc12aug_512x512_40k/model.pdparams) \| [log](https://bj.bcebos.com/paddleseg/dygraph/pascal_voc12/gcnet_resnet50_os8_voc12aug_512x512_40k/train.log) \| [vdl](https://paddlepaddle.org.cn/paddle/visualdl/service/app?id=86cbaac3fe98fdbb635e246c2c02e87b)|
+|GCNet|ResNet101_OS8|512x512|40000|79.64%|79.59%|79.94%|[model](https://bj.bcebos.com/paddleseg/dygraph/pascal_voc12/gcnet_resnet101_os8_voc12aug_512x512_40k/model.pdparams) \| [log](https://bj.bcebos.com/paddleseg/dygraph/pascal_voc12/gcnet_resnet101_os8_voc12aug_512x512_40k/train.log) \| [vdl](https://paddlepaddle.org.cn/paddle/visualdl/service/app?id=73f0484b034f6c27bf481c7a3b05e9ae)|
--- a/configs/gcnet/gcnet_resnet101_os8_cityscapes_1024x512_80k.yml
+++ b/configs/gcnet/gcnet_resnet101_os8_cityscapes_1024x512_80k.yml
+_base_: 'gcnet_resnet50_os8_cityscapes_1024x512_80k.yml'
+
+model:
+  backbone:
+    type: ResNet101_vd
+    pretrained: https://bj.bcebos.com/paddleseg/dygraph/resnet101_vd_ssld.tar.gz
--- a/configs/gcnet/gcnet_resnet101_os8_voc12aug_512x512_40k.yml
+++ b/configs/gcnet/gcnet_resnet101_os8_voc12aug_512x512_40k.yml
+_base_: './gcnet_resnet50_os8_voc12aug_512x512_40k.yml'
+
+model:
+  backbone:
+    type: ResNet101_vd
+    pretrained: https://bj.bcebos.com/paddleseg/dygraph/resnet101_vd_ssld.tar.gz
--- a/configs/gcnet/gcnet_resnet50_os8_cityscapes_1024x512_80k.yml
+++ b/configs/gcnet/gcnet_resnet50_os8_cityscapes_1024x512_80k.yml
+_base_: '../_base_/cityscapes.yml'
+
+batch_size: 2
+iters: 80000
+
+lr_scheduler:
+  type: PolynomialDecay
+  learning_rate: 0.01
+  power: 0.9
+  end_lr: 1.0e-5
+
+loss:
+  types:
+    - type: CrossEntropyLoss
+  coef: [1, 0.4]
+
+model:
+  type: GCNet
+  backbone:
+    type: ResNet50_vd
+    output_stride: 8
+    pretrained: https://bj.bcebos.com/paddleseg/dygraph/resnet50_vd_ssld_v2.tar.gz
+  gc_channels: 512
+  ratio: 0.25
+  enable_auxiliary_loss: True
+  align_corners: False
+  pretrained: null
--- a/configs/gcnet/gcnet_resnet50_os8_voc12aug_512x512_40k.yml
+++ b/configs/gcnet/gcnet_resnet50_os8_voc12aug_512x512_40k.yml
+_base_: '../_base_/pascal_voc12aug.yml'
+
+lr_scheduler:
+  type: PolynomialDecay
+  learning_rate: 0.01
+  power: 0.9
+  end_lr: 1.0e-5
+
+loss:
+  types:
+    - type: CrossEntropyLoss
+  coef: [1, 0.4]
+
+model:
+  type: GCNet
+  backbone:
+    type: ResNet50_vd
+    output_stride: 8
+    pretrained: https://bj.bcebos.com/paddleseg/dygraph/resnet50_vd_ssld_v2.tar.gz
+  gc_channels: 512
+  ratio: 0.25
+  enable_auxiliary_loss: True
+  align_corners: False
+  pretrained: null