Move Caffe models together.

examples/CaffeModels/README.md

+
+Example code to convert, load and run inference of some Caffe models.
+Require caffe python bindings to be installed.
+Converted models can also be found at [tensorpack model zoo](http://models.tensorpack.com).
+
+## AlexNet:
+
+Download: https://github.com/BVLC/caffe/tree/master/models/bvlc_alexnet
+
+Convert: `python -m tensorpack.utils.loadcaffe PATH/TO/CAFFE/{deploy.prototxt,bvlc_alexnet.caffemodel} alexnet.npz`
+
+Run: `./load-alexnet.py --load alexnet.npz --input cat.png`
+
+## VGG
+
+Download: https://gist.github.com/ksimonyan/211839e770f7b538e2d8
+
+Convert:
+```
+python -m tensorpack.utils.loadcaffe \
+            PATH/TO/VGG/{VGG_ILSVRC_16_layers_deploy.prototxt,VGG_ILSVRC_16_layers.caffemodel} vgg16.npz
+```
+
+Run: `./load-vgg16.py --load vgg16.npz --input cat.png`
+
+
+## ResNet
+
+To load caffe version of ResNet, see instructions in [ResNet examples](../ResNet).
+
+## Convolutional Pose Machines
+
+Download:
+```
+wget http://pearl.vasc.ri.cmu.edu/caffe_model_github/model/_trained_MPI/pose_iter_320000.caffemodel
+wget https://github.com/shihenw/convolutional-pose-machines-release/raw/master/model/_trained_MPI/pose_deploy_resize.prototxt
+```
+
+Convert: `python -m tensorpack.utils.loadcaffe pose_deploy_resize.prototxt pose_iter_320000.caffemodel CPM-original.npz`
+
+Run: `python load-cpm.py --load CPM-original.npz --input test.jpg`
+
+Input image will get resized to 368x368. Note that this CPM comes __without__ person detection, so the
+person has to be in the center of the image (and not too small).
+
+![demo](demo-cpm.jpg)
+
+Also check out [Stereo Pose Machines](https://github.com/ppwwyyxx/Stereo-Pose-Machines), a real-time CPM application based on tensorpack.

examples/load-alexnet.py → examples/CaffeModels/load-alexnet.py

@@ -15,20 +15,6 @@ from tensorpack.tfutils.summary import *
 from tensorpack.dataflow.dataset import ILSVRCMeta
 import tensorflow as tf
 
-"""
-Usage:
-    Download caffe models at https://github.com/BVLC/caffe/tree/master/models/bvlc_alexnet
-
-    Install caffe python bindings.
-
-    python -m tensorpack.utils.loadcaffe PATH/TO/CAFFE/{deploy.prototxt,bvlc_alexnet.caffemodel} alexnet.npz
-
-    Or download a converted caffe model from http://models.tensorpack.com/caffe/
-
-    Then, run it:
-    ./load-alexnet.py --load alexnet.npz --input cat.png
-"""
-
 
 def tower_func(image):
     # img: 227x227x3

examples/load-vgg16.py → examples/CaffeModels/load-vgg16.py

@@ -15,22 +15,6 @@ from tensorpack.tfutils.symbolic_functions import *
 from tensorpack.tfutils.summary import *
 from tensorpack.dataflow.dataset import ILSVRCMeta
 
-"""
-Usage:
-    Download original caffe models at:
-    https://gist.github.com/ksimonyan/211839e770f7b538e2d8
-
-    Install caffe python bindings.
-
-    python -m tensorpack.utils.loadcaffe \
-            PATH/TO/VGG/{VGG_ILSVRC_16_layers_deploy.prototxt,VGG_ILSVRC_16_layers.caffemodel} vgg16.npz
-
-    Or download a converted caffe model from http://models.tensorpack.com/caffe/
-
-    Then, run it:
-    ./load-vgg16.py --load vgg16.npz --input cat.png
-"""
-
 
 def tower_func(image):
     with argscope(Conv2D, kernel_shape=3, nl=tf.nn.relu):

examples/load-vgg19.py → examples/CaffeModels/load-vgg19.py

@@ -14,16 +14,6 @@ from tensorpack.tfutils.symbolic_functions import *
 from tensorpack.tfutils.summary import *
 from tensorpack.dataflow.dataset import ILSVRCMeta
 
-"""
-Usage:
-    python -m tensorpack.utils.loadcaffe \
-            PATH/TO/VGG/{VGG_ILSVRC_19_layers_deploy.prototxt,VGG_ILSVRC_16_layers.caffemodel} vgg19.npz
-    ./load-vgg19.py --load vgg19.npz --input cat.png
-
-    Or download a converted caffe model from http://models.tensorpack.com/caffe/
-    ./load-vgg19.py --load vgg19.npz --input cat.png
-"""
-
 
 def tower_func(image):
     with argscope(Conv2D, kernel_shape=3, nl=tf.nn.relu):

examples/ConvolutionalPoseMachines/README.md

-# Convolutional Pose Machines
-
-A script to load and run pre-trained CPM model released by Shih-En. The original code in caffe is [here](https://github.com/shihenw/convolutional-pose-machines-release).
-Reference paper: [Convolutional Pose Machines](https://arxiv.org/abs/1602.00134), Shih-En et al., CVPR16.
-
-Also check out [Stereo Pose Machines](https://github.com/ppwwyyxx/Stereo-Pose-Machines), a __real-time__ CPM application based on tensorpack.
-
-## Usage:
-
-Prepare the model:
-```
-# download the released caffe model:
-wget http://pearl.vasc.ri.cmu.edu/caffe_model_github/model/_trained_MPI/pose_iter_320000.caffemodel
-wget https://github.com/shihenw/convolutional-pose-machines-release/raw/master/model/_trained_MPI/pose_deploy_resize.prototxt
-# convert the model to a dict:
-python -m tensorpack.utils.loadcaffe pose_deploy_resize.prototxt pose_iter_320000.caffemodel CPM-original.npz
-```
-
-Or you can download the converted model from [model zoo](http://models.tensorpack.com/caffe/).
-
-Run it on an image, and produce `output.jpg`:
-```
-python load-cpm.py --load CPM-original.npz --input test.jpg
-```
-Input image will get resized to 368x368. Note that this CPM comes without person detection, so the
-person has to be in the center of the image (and not too small).
-
-![demo](demo.jpg)
+Moved to [../CaffeModels](../CaffeModels).

examples/ImageNetModels/inception-bn.py

@@ -22,6 +22,7 @@ NR_GPU = 6
 BATCH_SIZE = TOTAL_BATCH_SIZE // NR_GPU
 INPUT_SHAPE = 224
 
+
 class Model(ModelDesc):
     def _get_inputs(self):
         return [InputDesc(tf.float32, [None, INPUT_SHAPE, INPUT_SHAPE, 3], 'input'),

examples/ImageNetModels/vgg16.py

@@ -2,20 +2,13 @@
 # -*- coding: UTF-8 -*-
 # File: vgg16.py
 
-import sys
 import argparse
-import numpy as np
 import os
-from itertools import count
 
 import tensorflow as tf
 
 from tensorpack import *
-from tensorpack.models import *
-from tensorpack.callbacks import *
-from tensorpack.train import TrainConfig, SyncMultiGPUTrainerParameterServer
-from tensorpack.dataflow import imgaug
-from tensorpack.tfutils import argscope, get_model_loader, get_current_tower_context
+from tensorpack.tfutils import argscope, get_model_loader
 from tensorpack.tfutils.summary import *
 from tensorpack.utils.gpu import get_nr_gpu
 
@@ -37,7 +30,7 @@ class Model(ImageNetModel):
     def get_logits(self, image):
         with argscope(Conv2D, kernel_shape=3,
                       W_init=tf.variance_scaling_initializer(scale=2.)), \
-                  argscope([Conv2D, MaxPooling, BatchNorm], data_format='NCHW'):
+                argscope([Conv2D, MaxPooling, BatchNorm], data_format='NCHW'):
             logits = (LinearWrap(image)
                       .apply(convnormrelu, 'conv1_1', 64)
                       .apply(convnormrelu, 'conv1_2', 64)
@@ -63,18 +56,19 @@ class Model(ImageNetModel):
                       .MaxPooling('pool5', 2)
                       # 7
                       .FullyConnected('fc6', 4096,
-                          W_init=tf.random_normal_initializer(stddev=0.001))
+                                      W_init=tf.random_normal_initializer(stddev=0.001))
                       .tf.nn.relu(name='fc6_relu')
                       .Dropout('drop0', rate=0.5)
                       .FullyConnected('fc7', 4096,
-                          W_init=tf.random_normal_initializer(stddev=0.001))
+                                      W_init=tf.random_normal_initializer(stddev=0.001))
                       .tf.nn.relu(name='fc7_relu')
                       .Dropout('drop1', rate=0.5)
                       .FullyConnected('fc8', 1000,
-                          W_init=tf.random_normal_initializer(stddev=0.01))())
+                                      W_init=tf.random_normal_initializer(stddev=0.01))())
         add_param_summary(('.*', ['histogram', 'rms']))
         return logits
 
+
 def get_data(name, batch):
     isTrain = name == 'train'
     global args
@@ -98,9 +92,11 @@ def get_config():
         ModelSaver(),
         GPUUtilizationTracker(),
         EstimatedTimeLeft(),
-        ScheduledHyperParamSetter('learning_rate',
+        ScheduledHyperParamSetter(
+            'learning_rate',
             [(0, 0.01), (3, max(BASE_LR, 0.01))], interp='linear'),
-        ScheduledHyperParamSetter('learning_rate',
+        ScheduledHyperParamSetter(
+            'learning_rate',
             [(30, BASE_LR * 1e-1), (60, BASE_LR * 1e-2), (80, BASE_LR * 1e-3)]),
         DataParallelInferenceRunner(
             dataset_val, infs, list(range(nr_tower))),

examples/README.md

@@ -20,18 +20,17 @@ These examples don't have meaningful performance numbers. They are supposed to b
 ## Vision:
 | Name | Performance |
 | ---  | --- |
-|	Train [ResNet](ResNet) and [ShuffleNet](ShuffleNet) on ImageNet		| reproduce paper	|
+|	Train [ResNet](ResNet) and [ShuffleNet](ImageNetModels) on ImageNet		| reproduce paper	|
 |	[Train Faster-RCNN / Mask-RCNN on COCO](FasterRCNN)				|	reproduce paper		|
 | [DoReFa-Net: training binary / low-bitwidth CNN on ImageNet](DoReFa-Net) | reproduce paper |
 | [Generative Adversarial Network(GAN) variants](GAN), including DCGAN, InfoGAN, <br/> Conditional GAN, WGAN, BEGAN, DiscoGAN, Image to Image, CycleGAN | visually reproduce |
-| [Inception-BN](Inception) | reproduce reference code |
 | [Fully-convolutional Network for Holistically-Nested Edge Detection(HED)](HED) | visually reproduce |
 | [Spatial Transformer Networks on MNIST addition](SpatialTransformer) | reproduce paper |
 | [Visualize CNN saliency maps](Saliency) | visually reproduce |
 | [Similarity learning on MNIST](SimilarityLearning) | |
 | Single-image super-resolution using [EnhanceNet](SuperResolution) | visually reproduce |
 | Learn steering filters with [Dynamic Filter Networks](DynamicFilterNetwork) | visually reproduce |
-| Load a pre-trained [AlexNet](load-alexnet.py), [VGG16](load-vgg16.py), or [Convolutional Pose Machines](ConvolutionalPoseMachines/) | |
+| Load a pre-trained [AlexNet, VGG, or Convolutional Pose Machines](CaffeModels) | |
 
 ## Reinforcement Learning:
 | Name | Performance |

tests/dev/git-hooks/pre-commit

@@ -7,7 +7,7 @@ GIT_ARG="--git-dir ../.git --work-tree .."
 # find out modified python files, so that we ignored unstaged files
 # exclude ../docs
 MOD=$(git $GIT_ARG status -s | grep -E '\.py$'  \
-	| grep -E '^\b+M\b+' | cut -c 4- | grep -v '../docs')
+	| grep -E '^\b+M\b+|^A' | cut -c 4- | grep -v '../docs')
 if [[ -n $MOD ]]; then
 	flake8 $MOD
 fi