文件列表:

assignment5-modified codes\__pycache__ (0, 2020-06-30)
assignment5-modified codes\__pycache__\conviz.cpython-36.pyc (2274, 2018-12-14)
assignment5-modified codes\__pycache__\mymodel.cpython-36.pyc (4318, 2018-12-14)
assignment5-modified codes\conviz.py (3139, 2018-12-14)
assignment5-modified codes\datasets (0, 2020-06-30)
assignment5-modified codes\datasets\__init__.py (1, 2018-12-14)
assignment5-modified codes\datasets\__pycache__ (0, 2020-06-30)
assignment5-modified codes\datasets\__pycache__\__init__.cpython-36.pyc (140, 2018-12-14)
assignment5-modified codes\datasets\__pycache__\cifar10.cpython-36.pyc (2263, 2018-12-14)
assignment5-modified codes\datasets\__pycache__\dataset_utils.cpython-36.pyc (4373, 2018-12-14)
assignment5-modified codes\datasets\cifar10.py (3237, 2018-12-14)
assignment5-modified codes\datasets\dataset_utils.py (4680, 2018-12-14)
assignment5-modified codes\images (0, 2020-06-30)
assignment5-modified codes\images\conv_output (0, 2020-06-30)
assignment5-modified codes\images\conv_output\layer11 (0, 2020-06-30)
assignment5-modified codes\images\conv_output\layer11\layer11.png (16627, 2018-12-14)
assignment5-modified codes\images\conv_output\layer5 (0, 2020-06-30)
assignment5-modified codes\images\conv_output\layer5\layer5.png (51160, 2018-12-14)
assignment5-modified codes\images\conv_weights (0, 2020-06-30)
assignment5-modified codes\images\conv_weights\layer11 (0, 2020-06-30)
assignment5-modified codes\images\conv_weights\layer11\layer11-33.png (10003, 2018-12-14)
assignment5-modified codes\images\conv_weights\layer5 (0, 2020-06-30)
assignment5-modified codes\images\conv_weights\layer5\layer5-0.png (9978, 2018-12-14)
assignment5-modified codes\model_train.py (5947, 2018-12-14)
assignment5-modified codes\model_visualize.py (2133, 2018-12-14)
assignment5-modified codes\mymodel.py (5558, 2018-12-14)
assignment5-modified codes\visual (0, 2020-06-30)
assignment5-modified codes\visual\__init__.py (0, 2018-12-14)
assignment5-modified codes\visual\__pycache__ (0, 2020-06-30)
assignment5-modified codes\visual\__pycache__\__init__.cpython-36.pyc (138, 2018-12-14)
assignment5-modified codes\visual\__pycache__\conviz.cpython-36.pyc (2281, 2018-12-14)
assignment5-modified codes\visual\__pycache__\utils.cpython-36.pyc (2107, 2018-12-14)
assignment5-modified codes\visual\utils.py (1805, 2018-12-14)

# Build own model using tensorflow api and filters and layers visualization Let's say, the structure of the model is: | Input (32 x 32 RGB images) | Layers| |:----------:|:-------:| | Conv3-8 | Layer-1 | | Conv3-8 | Layer-2 | | Conv3-8 | Layer-3 | | maxpool | Layer-4 | | Conv3-*** | Layer-5 | | Conv3-*** | Layer-6 | | Conv3-*** | Layer-7 | | maxpool | Layer-8 | | Conv3-*** | Layer-9 | | Conv3-*** | Layer-10 | | Conv3-*** | Layer-11 | | maxpool | Layer-12 | | FC-1024 | Layer-13 | | FC-10 | Layer-14 | | softmax | Layer-15 | * Conv3-8 means the convolutional kernel is 3 — 3, and number of output channels is 8, the padding style is **SAME** rather than **VALID** (see `tf.nn.conv2d`) * FC-1024 means the output size of the FC layer is 1024 * stride of Conv layers is 1, stride of pooling layers is 2 * kernel size of maxpool is 2 x 2 ## Dirs used Here, we spicify the following paths for the model for convenience. Please change them to your own path when doing yourself.
**Path where you download cifar10 dataset:** */mymodel/cifar10-data*
**Path where you save your model to when training and load your model from when testing or finetuning:** */mymodel/model*
**Path where visualization results are saved:** */mymodel/visual_results*
# Preparation Refer to codes in assignment4 for training and using tensorboard ## visualization Refer to `model_visualize.py` or [the original github repository](https://github.com/grishasergei/conviz)
Run example: First, add your code at the end of `model_visualize.py`
```bash python model_visualize.py \ --train_dir='/mymodel/model' \ --data_dir='/mymodel/cifar10_data' \ --visual_dir='/mymodel/visual_results' ``` The results of visualization are like this
**con_weights_layer5_channel_0** (show only image of the first channel)
 **conv_output_layer5**
 **con_weights_layer11_channel_33** (image of channel 34)
 **conv_output_layer11**
 Any problem: email zhuo.su@oulu.fi

近期下载者：

相关文件：

评论：[我要评论] [举报此文件]

收藏者：