Improve resnet18 and resnet34 pre-trained models (#10752)

* fix #9865 * add unittest * fix format * fix format * fix superfluous loop in metric * fix lint * update resnet18/34 in the gluon model zoo * correct values
apache · May 1, 2018 · 8585ef5 · 8585ef5
1 parent 61f86fc
commit 8585ef5
Show file tree

Hide file tree

Showing 2 changed files with 8 additions and 8 deletions.
diff --git a/docs/api/python/gluon/model_zoo.md b/docs/api/python/gluon/model_zoo.md
@@ -42,13 +42,13 @@ The following table summarizes the available models.
 | mobilenet0.5  | [MobileNet 0.5](https://arxiv.org/abs/1704.04861)                                     | 1,342,536    | 0.6307         | 0.8475         | Trained with [script](https://github.com/apache/incubator-mxnet/blob/master/example/gluon/image_classification.py)              |
 | mobilenet0.75 | [MobileNet 0.75](https://arxiv.org/abs/1704.04861)                                    | 2,601,976    | 0.6738         | 0.8782         | Trained with [script](https://github.com/apache/incubator-mxnet/blob/master/example/gluon/image_classification.py)              |
 | mobilenet1.0  | [MobileNet 1.0](https://arxiv.org/abs/1704.04861)                                     | 4,253,864    | 0.7105         | 0.9006         | Trained with [script](https://github.com/apache/incubator-mxnet/blob/master/example/gluon/image_classification.py)              |
-| resnet18_v1   | [ResNet-18 V1](http://arxiv.org/abs/1512.03385)                                       | 11,699,112   | 0.6803         | 0.8818         | Converted from pytorch vision                                                                                                                        |
-| resnet34_v1   | [ResNet-34 V1](http://arxiv.org/abs/1512.03385)                                       | 21,814,696   | 0.7202         | 0.9066         | Converted from pytorch vision                                                                                                                        |
+| resnet18_v1   | [ResNet-18 V1](http://arxiv.org/abs/1512.03385)                                       | 11,699,112   | 0.7039         | 0.8959         | Trained with [script](https://github.com/dmlc/gluon-cv/blob/15ed8a4c71d411b878f0d71d1c7afdce6710c913/scripts/classification/imagenet/train_imagenet.py) |
+| resnet34_v1   | [ResNet-34 V1](http://arxiv.org/abs/1512.03385)                                       | 21,814,696   | 0.7411         | 0.9184         | Trained with [script](https://github.com/dmlc/gluon-cv/blob/15ed8a4c71d411b878f0d71d1c7afdce6710c913/scripts/classification/imagenet/train_imagenet.py) |
 | resnet50_v1   | [ResNet-50 V1](http://arxiv.org/abs/1512.03385)                                       | 25,629,032   | 0.7540         | 0.9266         | Trained with [script](https://github.com/apache/incubator-mxnet/blob/master/example/gluon/image_classification.py)              |
 | resnet101_v1  | [ResNet-101 V1](http://arxiv.org/abs/1512.03385)                                      | 44,695,144   | 0.7693         | 0.9334         | Trained with [script](https://github.com/apache/incubator-mxnet/blob/master/example/gluon/image_classification.py)              |
 | resnet152_v1  | [ResNet-152 V1](http://arxiv.org/abs/1512.03385)                                      | 60,404,072   | 0.7727         | 0.9353         | Trained with [script](https://github.com/apache/incubator-mxnet/blob/master/example/gluon/image_classification.py)              |
-| resnet18_v2   | [ResNet-18 V2](https://arxiv.org/abs/1603.05027)                                      | 11,695,796   | 0.6961         | 0.8901         | Trained with [script](https://github.com/apache/incubator-mxnet/blob/4dcd96ae2f6820e01455079d00f49db1cd21eda9/example/gluon/image_classification.py) |
-| resnet34_v2   | [ResNet-34 V2](https://arxiv.org/abs/1603.05027)                                      | 21,811,380   | 0.7324         | 0.9125         | Trained with [script](https://github.com/apache/incubator-mxnet/blob/4dcd96ae2f6820e01455079d00f49db1cd21eda9/example/gluon/image_classification.py) |
+| resnet18_v2   | [ResNet-18 V2](https://arxiv.org/abs/1603.05027)                                      | 11,695,796   | 0.7037         | 0.8962         | Trained with [script](https://github.com/dmlc/gluon-cv/blob/15ed8a4c71d411b878f0d71d1c7afdce6710c913/scripts/classification/imagenet/train_imagenet.py) |
+| resnet34_v2   | [ResNet-34 V2](https://arxiv.org/abs/1603.05027)                                      | 21,811,380   | 0.7389         | 0.9178         | Trained with [script](https://github.com/dmlc/gluon-cv/blob/15ed8a4c71d411b878f0d71d1c7afdce6710c913/scripts/classification/imagenet/train_imagenet.py) |
 | resnet50_v2   | [ResNet-50 V2](https://arxiv.org/abs/1603.05027)                                      | 25,595,060   | 0.7622         | 0.9297         | Trained with [script](https://github.com/apache/incubator-mxnet/blob/master/example/gluon/image_classification.py)              |
 | resnet101_v2  | [ResNet-101 V2](https://arxiv.org/abs/1603.05027)                                     | 44,639,412   | 0.7747         | 0.9375         | Trained with [script](https://github.com/apache/incubator-mxnet/blob/master/example/gluon/image_classification.py)              |
 | resnet152_v2  | [ResNet-152 V2](https://arxiv.org/abs/1603.05027)                                     | 60,329,140   | 0.7833         | 0.9409         | Trained with [script](https://github.com/apache/incubator-mxnet/blob/master/example/gluon/image_classification.py)              |

diff --git a/python/mxnet/gluon/model_zoo/model_store.py b/python/mxnet/gluon/model_zoo/model_store.py
@@ -35,13 +35,13 @@
     ('8e9d539cc66aa5efa71c4b6af983b936ab8701c3', 'mobilenet0.5'),
     ('529b2c7f4934e6cb851155b22c96c9ab0a7c4dc2', 'mobilenet0.75'),
     ('6b8c5106c730e8750bcd82ceb75220a3351157cd', 'mobilenet1.0'),
-    ('38d6d423c22828718ec3397924b8e116a03e6ac0', 'resnet18_v1'),
-    ('4dc2c2390a7c7990e0ca1e53aeebb1d1a08592d1', 'resnet34_v1'),
+    ('e54b379f50fa4b10bbd2506237e3bd74e6164778', 'resnet18_v1'),
+    ('c1dc0967a3d25ee9127e03bc1046a5d44d92e2ba', 'resnet34_v1'),
     ('c940b1a062b32e3a5762f397c9d1e178b5abd007', 'resnet50_v1'),
     ('d992389084bc5475c370e9b52c3561706e755799', 'resnet101_v1'),
     ('48ce7775d375987d019ec9aa96bc43b98165dfcb', 'resnet152_v1'),
-    ('8aacf80ff4014c1efa2362a963ac5ec82cf92d5b', 'resnet18_v2'),
-    ('0ed3cd06da41932c03dea1de7bc2506ef3fb97b3', 'resnet34_v2'),
+    ('84f666402577b5b79cd59eba5d3ba0bc1edf2152', 'resnet18_v2'),
+    ('5da34c2772893e9d680d5fa0bd6d432eba8689c9', 'resnet34_v2'),
     ('81a4e66af7859a5aa904e2b4051aa0d3bc472b2f', 'resnet50_v2'),
     ('7eb2b3cde097883c11941b927048a705ed334294', 'resnet101_v2'),
     ('64c75ac8c292f6ac54f873f9ef62e0531105878b', 'resnet152_v2'),