update readme and pretrain path

configs/fast_scnn_pet.yaml

@@ -31,7 +31,7 @@ MODEL:
     DEFAULT_NORM_TYPE: "bn"
 
 TRAIN:
-    PRETRAINED_MODEL_DIR: "./pretrained_model/fast_scnn_cityscape/"
+    PRETRAINED_MODEL_DIR: "./pretrained_model/fast_scnn_cityscapes/"
     MODEL_SAVE_DIR: "./saved_model/fast_scnn_pet/"
     SNAPSHOT_EPOCH: 10
 TEST:

turtorial/finetune_fast_scnn.md

@@ -68,7 +68,7 @@ AUG:
     FIX_RESIZE_SIZE: (512, 512)
 BATCH_SIZE: 4
 TRAIN:
-    PRETRAINED_MODEL_DIR: "./pretrained_model/fast_scnn_cityscape/"
+    PRETRAINED_MODEL_DIR: "./pretrained_model/fast_scnn_cityscapes/"
     MODEL_SAVE_DIR: "./saved_model/fast_scnn_pet/"
     SNAPSHOT_EPOCH: 10
 TEST:
@@ -114,6 +114,6 @@ python pdseg/eval.py --use_gpu --cfg ./configs/fast_scnn_pet.yaml
 | ICNet/bn |(1024, 2048) |8.76ms| 0.6831 |
 | Fast-SCNN/bn | (1024, 2048) |6.28ms| 0.6964 |
 
-上述测试环境为v100.
+上述测试环境为v100. 测试使用paddle的推理接口[zero_copy]()的方式，模型输出是类别，即argmax后的类别。