From 4c3effee83cad0e32a1c8ec6b4c484a21f292037 Mon Sep 17 00:00:00 2001
From: LielinJiang <jianglielin@baidu.com>
Date: Tue, 21 Apr 2020 11:30:57 +0000
Subject: [PATCH] add darknet params

---
 examples/image_classification/README.MD       |  3 ++-
 .../image_classification/imagenet_dataset.py  | 11 +++++++---
 examples/image_classification/main.py         | 21 +++++++++++++++----
 hapi/vision/models/darknet.py                 |  5 +++--
 4 files changed, 30 insertions(+), 10 deletions(-)

diff --git a/examples/image_classification/README.MD b/examples/image_classification/README.MD
index 5b50370..9f2c585 100644
--- a/examples/image_classification/README.MD
+++ b/examples/image_classification/README.MD
@@ -85,8 +85,9 @@ CUDA_VISIBLE_DEVICES=0,1,2,3 python -m paddle.distributed.launch main.py --arch
 | [vgg16](https://paddle-hapi.bj.bcebos.com/models/vgg16.pdparams) | 71.92 | 90.65 | 
 | [mobilenet_v1](https://paddle-hapi.bj.bcebos.com/models/mobilenet_v1_x1.0.pdparams) | 71.16 | 89.89 | 
 | [mobilenet_v2](https://paddle-hapi.bj.bcebos.com/models/mobilenet_v2_x1.0.pdparams) | 72.30 | 90.74 | 
+| [darknet53](https://paddle-hapi.bj.bcebos.com/models/darknet53.pdparams) | 78.43 | 94.24 | 
 
-上述模型的复现参数请参考scripts下的脚本。
+上述部分模型的复现参数请参考scripts下的脚本。需要注意的是darknet要使用image size为256的输入来预测, 即```--image-size 256```
 
 
 ## 参考文献
diff --git a/examples/image_classification/imagenet_dataset.py b/examples/image_classification/imagenet_dataset.py
index 6572df0..3d730de 100644
--- a/examples/image_classification/imagenet_dataset.py
+++ b/examples/image_classification/imagenet_dataset.py
@@ -24,7 +24,11 @@ from paddle import fluid
 
 
 class ImageNetDataset(DatasetFolder):
-    def __init__(self, path, mode='train'):
+    def __init__(self,
+                 path,
+                 mode='train',
+                 image_size=224,
+                 resize_short_size=256):
         super(ImageNetDataset, self).__init__(path)
         self.mode = mode
 
@@ -32,13 +36,14 @@ class ImageNetDataset(DatasetFolder):
             mean=[123.675, 116.28, 103.53], std=[58.395, 57.120, 57.375])
         if self.mode == 'train':
             self.transform = transforms.Compose([
-                transforms.RandomResizedCrop(224),
+                transforms.RandomResizedCrop(image_size),
                 transforms.RandomHorizontalFlip(),
                 transforms.Permute(mode='CHW'), normalize
             ])
         else:
             self.transform = transforms.Compose([
-                transforms.Resize(256), transforms.CenterCrop(224),
+                transforms.Resize(resize_short_size),
+                transforms.CenterCrop(image_size),
                 transforms.Permute(mode='CHW'), normalize
             ])
 
diff --git a/examples/image_classification/main.py b/examples/image_classification/main.py
index 76360df..da47ca4 100644
--- a/examples/image_classification/main.py
+++ b/examples/image_classification/main.py
@@ -18,8 +18,6 @@ from __future__ import print_function
 import argparse
 import contextlib
 import os
-import sys
-sys.path.append('../')
 
 import time
 import math
@@ -89,8 +87,16 @@ def main():
     labels = [Input([None, 1], 'int64', name='label')]
 
     train_dataset = ImageNetDataset(
-        os.path.join(FLAGS.data, 'train'), mode='train')
-    val_dataset = ImageNetDataset(os.path.join(FLAGS.data, 'val'), mode='val')
+        os.path.join(FLAGS.data, 'train'),
+        mode='train',
+        image_size=FLAGS.image_size,
+        resize_short_size=FLAGS.resize_short_size)
+
+    val_dataset = ImageNetDataset(
+        os.path.join(FLAGS.data, 'val'),
+        mode='val',
+        image_size=FLAGS.image_size,
+        resize_short_size=FLAGS.resize_short_size)
 
     optim = make_optimizer(
         np.ceil(
@@ -176,6 +182,13 @@ if __name__ == '__main__':
     parser.add_argument(
         "--weight-decay", default=1e-4, type=float, help="weight decay")
     parser.add_argument("--momentum", default=0.9, type=float, help="momentum")
+    parser.add_argument(
+        "--image-size", default=224, type=int, help="intput image size")
+    parser.add_argument(
+        "--resize-short-size",
+        default=256,
+        type=float,
+        help="short size of keeping ratio resize")
     FLAGS = parser.parse_args()
     assert FLAGS.data, "error: must provide data path"
     main()
diff --git a/hapi/vision/models/darknet.py b/hapi/vision/models/darknet.py
index 42f6a2e..f1d8e03 100755
--- a/hapi/vision/models/darknet.py
+++ b/hapi/vision/models/darknet.py
@@ -26,8 +26,8 @@ __all__ = ['DarkNet', 'ConvBNLayer', 'darknet53']
 
 # {num_layers: (url, md5)}
 pretrain_infos = {
-    53: ('https://paddlemodels.bj.bcebos.com/hapi/darknet53.pdparams',
-         '2506357a5c31e865785112fc614a487d')
+    53: ('https://paddle-hapi.bj.bcebos.com/models/darknet53.pdparams',
+         'ca506a90e2efecb9a2093f8ada808708')
 }
 
 
@@ -67,6 +67,7 @@ class ConvBNLayer(fluid.dygraph.Layer):
     def forward(self, inputs):
         out = self.conv(inputs)
         out = self.batch_norm(out)
+        # out = fluid.layers.relu(out)
         if self.act == 'leaky':
             out = fluid.layers.leaky_relu(x=out, alpha=0.1)
         return out
-- 
GitLab