Add Inception-V4 for Fluid #672
Conversation
… models_issues_647 Add Inception-V4 for Fluid PaddlePaddle#647
| import numpy | ||
| import paddle.v2 as paddle | ||
| import paddle.fluid as fluid | ||
| import math |
There was a problem hiding this comment.
line 20 移动到 line 16行之后。在 import paddle.v2 之前空一行 ：https://www.python.org/dev/peps/pep-0008/#imports
| import math | ||
|
|
||
| __all__ = ['inception_v4'] | ||
|
|
There was a problem hiding this comment.
22 行移动到 15 行之后，并且空一行：https://www.python.org/dev/peps/pep-0008/#module-level-dunder-names
| input=input, | ||
| pool_size=3, | ||
| pool_stride=2, | ||
| # add |
There was a problem hiding this comment.
line 50 行是否有特殊含义，如果没有则删除。
| input=input, num_filters=96, filter_size=3, stride=2, padding=1) | ||
| return fluid.layers.concat(input=[pool0, conv0], axis=1) | ||
|
|
||
| def block1(input): |
There was a problem hiding this comment.
请为每一个 python函数添加docstring注释。风格可以参考 https://www.python.org/dev/peps/pep-0008/#documentation-strings
| image = fluid.layers.data(name='image', shape=image_shape, dtype='float32') | ||
| label = fluid.layers.data(name='label', shape=[1], dtype='int64') | ||
|
|
||
| net = inception_v4(image, class_dim) |
There was a problem hiding this comment.
| sys.exit("got NaN loss, training failed.") | ||
| acc_list.append(float(acc_t)) | ||
| avg_loss_list.append(float(loss_t)) | ||
| break # Use 1 segment for speeding up CI |
There was a problem hiding this comment.
line 357 在models下不需要，models下都是可以完整运行的例子。请删除这一行的逻辑。
| fluid.io.save_inference_model(model_save_dir, ["image"], | ||
| [predict], exe) | ||
| return | ||
|
|
There was a problem hiding this comment.
models 下都是可以直接运行的完整例子，line 367 ~ 370 不需要这部分。
|
对models下代码merge要求是：需要保证收敛和学习效果，但不以对齐到论文state-of-art为merge的强制要求。请问PR中的 Inception-V4 模型收敛效果如何呢？能否在comment中提供训练效果数据呢？ |
|
|
resolve #647