提交 d7cac6e8 编写于 作者: L LielinJiang

update readme and pretrain path

上级 93965164
......@@ -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:
......
......@@ -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。
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册