[Feature] Add ENet

configs/enet/README.md

@@ -0,0 +1,13 @@
+# ENet: A Deep Neural Network Architecture forReal-Time Semantic Segmentation
+
+## Reference
+> Abhishek Chaurasia, Sangpil Kim, Eugenio Culurciello. "ENet: A Deep Neural Network Architecture for
+Real-Time Semantic Segmentation." arXiv preprint arXiv:1606.02147(2016).
+
+## Performance
+
+### Cityscapes
+
+| Model | Backbone | Resolution | Training Iters | mIoU | Links |
+|:-:|:-:|:-:|:-:|:-:|:-:|
+|ENet|-|1024x512|80000|61.3%|[提取码：2fle](https://pan.baidu.com/share/init?surl=7k-Mq_BmZp0FaY1IDjAVoQ)|

configs/enet/enet_cityscapes_1024x512_adam_0.002_80k.yml

@@ -0,0 +1,25 @@
+_base_: '../_base_/cityscapes.yml'
+
+batch_size: 8
+
+train_dataset:
+  type: Cityscapes
+  dataset_root: data/cityscapes
+  transforms:
+    - type: RandomPaddingCrop
+      crop_size: [1024, 512]
+    - type: RandomDistort
+      brightness_range: 0.4
+      contrast_range: 0.4
+      saturation_range: 0.4
+    - type: Normalize
+  mode: train
+
+model:
+  type: ENet
+  num_classes: 19
+  pretrained: Null
+
+optimizer:
+  type: adam
+  weight_decay: 0.0002

paddleseg/models/__init__.py

@@ -48,3 +48,4 @@
 from .segnet import SegNet
 from .hrnet_contrast import HRNetW48Contrast
 from .espnet import ESPNetV2
+from .enet import ENet