File movings (#9899)

* Dataset utilities added.

* Global model definition

* Dataset modules added.

* Dataset modules fix.

* global features model training added

* global features fix

* Test dataset update

* PR fixes

* repo sync

* repo sync

* modules moving

* Removed unnecessary modules

* Removed unnecessary files

research/delf/delf/__init__.py

@@ -35,5 +35,7 @@ from delf.python.examples import extractor
 from delf.python import detect_to_retrieve
 from delf.python import training
 from delf.python.training import model
-from delf.python.training import datasets
+from delf.python import datasets
+from delf.python.datasets import google_landmarks_dataset
+from delf.python.datasets import revisited_op
 # pylint: enable=unused-import

research/delf/delf/python/google_landmarks_dataset/README.md → research/delf/delf/python/datasets/google_landmarks_dataset/README.md

@@ -18,7 +18,7 @@ For more details on the dataset, please refer to its
 ### Install DELF library
 
 To be able to use this code, please follow
-[these instructions](../../../INSTALL_INSTRUCTIONS.md) to properly install the
+[these instructions](../../../../INSTALL_INSTRUCTIONS.md) to properly install the
 DELF library.
 
 ### Download Revisited Oxford/Paris datasets
@@ -48,7 +48,7 @@ wget http://cmp.felk.cvut.cz/revisitop/data/datasets/rparis6k/gnd_rparis6k.mat
 ### Download model
 
 ```bash
-# From models/research/delf/delf/python/google_landmarks_dataset
+# From models/research/delf/delf/python/datasets/google_landmarks_dataset
 mkdir parameters && cd parameters
 
 # RN101-ArcFace model trained on GLDv2-clean.
@@ -72,8 +72,8 @@ datasets, and not required for the GLDv2 retrieval/recognition datasets.
 Run query feature extraction as follows:
 
 ```bash
-# From models/research/delf/delf/python/google_landmarks_dataset
-python3 ../delg/extract_features.py \
+# From models/research/delf/delf/python/datasets/google_landmarks_dataset
+python3 ../../delg/extract_features.py \
   --delf_config_path rn101_af_gldv2clean_config.pbtxt \
   --dataset_file_path ~/revisitop/data/gnd_roxford5k.mat \
   --images_dir ~/revisitop/data/oxford5k_images \
@@ -86,8 +86,8 @@ python3 ../delg/extract_features.py \
 Run index feature extraction as follows:
 
 ```bash
-# From models/research/delf/delf/python/google_landmarks_dataset
-python3 ../delg/extract_features.py \
+# From models/research/delf/delf/python/datasets/google_landmarks_dataset
+python3 ../../delg/extract_features.py \
   --delf_config_path rn101_af_gldv2clean_config.pbtxt \
   --dataset_file_path ~/revisitop/data/gnd_roxford5k.mat \
   --images_dir ~/revisitop/data/oxford5k_images \
@@ -100,8 +100,8 @@ python3 ../delg/extract_features.py \
 To run retrieval on `roxford5k`, the following command can be used:
 
 ```bash
-# From models/research/delf/delf/python/google_landmarks_dataset
-python3 ../delg/perform_retrieval.py \
+# From models/research/delf/delf/python/datasets/google_landmarks_dataset
+python3 ../../delg/perform_retrieval.py \
   --dataset_file_path ~/revisitop/data/gnd_roxford5k.mat \
   --query_features_dir ~/revisitop/data/oxford5k_features/query \
   --index_features_dir ~/revisitop/data/oxford5k_features/index \

research/delf/delf/python/datasets/google_landmarks_dataset/__init__.py

+# Copyright 2021 The TensorFlow Authors All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+"""Module exposing Google Landmarks dataset for training."""
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+# pylint: disable=unused-import
+from delf.python.datasets.google_landmarks_dataset import googlelandmarks
+# pylint: enable=unused-import

research/delf/delf/python/google_landmarks_dataset/compute_recognition_metrics.py → research/delf/delf/python/datasets/google_landmarks_dataset/compute_recognition_metrics.py

@@ -25,8 +25,8 @@ import argparse
 import sys
 
 from tensorflow.python.platform import app
-from delf.python.google_landmarks_dataset import dataset_file_io
-from delf.python.google_landmarks_dataset import metrics
+from delf.python.datasets.google_landmarks_dataset import dataset_file_io
+from delf.python.datasets.google_landmarks_dataset import metrics
 
 cmd_args = None
 

research/delf/delf/python/google_landmarks_dataset/compute_retrieval_metrics.py → research/delf/delf/python/datasets/google_landmarks_dataset/compute_retrieval_metrics.py

@@ -25,8 +25,8 @@ import argparse
 import sys
 
 from tensorflow.python.platform import app
-from delf.python.google_landmarks_dataset import dataset_file_io
-from delf.python.google_landmarks_dataset import metrics
+from delf.python.datasets.google_landmarks_dataset import dataset_file_io
+from delf.python.datasets.google_landmarks_dataset import metrics
 
 cmd_args = None
 

research/delf/delf/python/google_landmarks_dataset/dataset_file_io_test.py → research/delf/delf/python/datasets/google_landmarks_dataset/dataset_file_io_test.py

@@ -23,7 +23,7 @@ import os
 from absl import flags
 import tensorflow as tf
 
-from delf.python.google_landmarks_dataset import dataset_file_io
+from delf.python.datasets.google_landmarks_dataset import dataset_file_io
 
 FLAGS = flags.FLAGS
 

research/delf/delf/python/google_landmarks_dataset/metrics_test.py → research/delf/delf/python/datasets/google_landmarks_dataset/metrics_test.py

@@ -20,7 +20,7 @@ from __future__ import print_function
 
 import tensorflow as tf
 
-from delf.python.google_landmarks_dataset import metrics
+from delf.python.datasets.google_landmarks_dataset import metrics
 
 
 def _CreateRecognitionSolution():

research/delf/delf/python/training/datasets/__init__.py → research/delf/delf/python/datasets/revisited_op/__init__.py

-# Copyright 2020 The TensorFlow Authors All Rights Reserved.
+# Copyright 2021 The TensorFlow Authors All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -12,11 +12,11 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 # ==============================================================================
-"""Module exposing datasets for training."""
+"""Module for revisited Oxford and Paris datasets."""
 from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
 # pylint: disable=unused-import
-from delf.python.training.datasets import googlelandmarks
+from delf.python.datasets.revisited_op import dataset
 # pylint: enable=unused-import

research/delf/delf/python/detect_to_retrieve/dataset.py → research/delf/delf/python/datasets/revisited_op/dataset.py

@@ -18,6 +18,9 @@ from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
 
+import os
+import pickle
+
 import numpy as np
 from scipy.io import matlab
 import tensorflow as tf
@@ -467,3 +470,65 @@ def ReadMetricsFile(metrics_path):
       dtype=float) / 100.0
 
   return mean_average_precision, pr_ranks, mean_precisions, mean_recalls
+
+
+def create_config_for_test_dataset(dataset, dir_main):
+  """Creates the configuration dictionary for the test dataset.
+
+  Args:
+    dataset: String, dataset name: either 'roxford5k' or 'rparis6k'.
+    dir_main: String, path to the folder containing ground truth files.
+
+  Returns:
+    cfg: Dataset configuration in a form of dictionary. The configuration
+      includes:
+      `gnd_fname` - path to the ground truth file for teh dataset,
+      `ext` and `qext` - image extentions for the images in the test dataset
+      and the query images,
+      `dir_data` - path to the folder containing ground truth files,
+      `dir_images` - path to the folder containing images,
+      `n` and `nq` - number of images and query images in the dataset
+      respectively,
+      `im_fname` and `qim_fname` - functions providing paths for the dataset
+      and query images respectively,
+      `dataset` - test dataset name.
+
+  Raises:
+    ValueError: If an unknown dataset name is provided as an argument.
+  """
+  DATASETS = ['roxford5k', 'rparis6k']
+  dataset = dataset.lower()
+
+  def _config_imname(cfg, i):
+    return os.path.join(cfg['dir_images'], cfg['imlist'][i] + cfg['ext'])
+
+  def _config_qimname(cfg, i):
+    return os.path.join(cfg['dir_images'], cfg['qimlist'][i] + cfg['qext'])
+
+  if dataset not in DATASETS:
+    raise ValueError('Unknown dataset: {}!'.format(dataset))
+
+  # Loading imlist, qimlist, and gnd in configuration as a dictionary.
+  gnd_fname = os.path.join(dir_main, 'gnd_{}.pkl'.format(dataset))
+  with tf.io.gfile.GFile(gnd_fname, 'rb') as f:
+    cfg = pickle.load(f)
+  cfg['gnd_fname'] = gnd_fname
+  if dataset == 'rparis6k':
+    dir_images = 'paris6k_images'
+  elif dataset == 'roxford5k':
+    dir_images = 'oxford5k_images'
+
+  cfg['ext'] = '.jpg'
+  cfg['qext'] = '.jpg'
+  cfg['dir_data'] = os.path.join(dir_main)
+  cfg['dir_images'] = os.path.join(cfg['dir_data'], dir_images)
+
+  cfg['n'] = len(cfg['imlist'])
+  cfg['nq'] = len(cfg['qimlist'])
+
+  cfg['im_fname'] = _config_imname
+  cfg['qim_fname'] = _config_qimname
+
+  cfg['dataset'] = dataset
+
+  return cfg

research/delf/delf/python/datasets/utils.py

@@ -22,14 +22,15 @@ import tensorflow as tf
 from delf.python import utils as image_loading_utils
 
 
-def pil_imagenet_loader(path, imsize, bounding_box=None, normalize=True):
+def pil_imagenet_loader(path, imsize, bounding_box=None, preprocess=True):
   """Pillow loader for the images.
 
   Args:
     path: Path to image to be loaded.
     imsize: Integer, defines the maximum size of longer image side.
     bounding_box: (x1,y1,x2,y2) tuple to crop the query image.
-    normalize: Bool, whether to normalize the image.
+    preprocess: Bool, whether to preprocess the images in respect to the
+      ImageNet dataset.
 
   Returns: 
     image: `Tensor`, image in ImageNet suitable format.
@@ -47,7 +48,7 @@ def pil_imagenet_loader(path, imsize, bounding_box=None, normalize=True):
   img.thumbnail((imsize, imsize), Image.ANTIALIAS)
   img = np.array(img)
 
-  if normalize:
+  if preprocess:
     # Preprocessing for ImageNet data. Converts the images from RGB to BGR,
     # then zero-centers each color channel with respect to the ImageNet
     # dataset, without scaling.
@@ -56,16 +57,18 @@ def pil_imagenet_loader(path, imsize, bounding_box=None, normalize=True):
   return img
 
 
-def default_loader(path, imsize, bounding_box=None, normalize=True):
+def default_loader(path, imsize, bounding_box=None, preprocess=True):
   """Default loader for the images is using Pillow.
 
   Args:
     path: Path to image to be loaded.
     imsize: Integer, defines the maximum size of longer image side.
     bounding_box: (x1,y1,x2,y2) tuple to crop the query image.
+    preprocess: Bool, whether to preprocess the images in respect to the
+      ImageNet dataset.
 
   Returns:
     image: `Tensor`, image in ImageNet suitable format.
   """
-  img = pil_imagenet_loader(path, imsize, bounding_box, normalize)
+  img = pil_imagenet_loader(path, imsize, bounding_box, preprocess)
   return img

research/delf/delf/python/datasets/utils_test.py

@@ -44,7 +44,7 @@ class UtilsTest(tf.test.TestCase):
     max_img_size = 1024
     # Load the saved dummy image.
     img = image_loading_utils.default_loader(filename, imsize=max_img_size,
-                                     normalize=False)
+                                             preprocess=False)
 
     # Make sure the values are the same before and after loading.
     self.assertAllEqual(np.array(img_out), img)
@@ -62,13 +62,12 @@ class UtilsTest(tf.test.TestCase):
     max_img_size = 1024
     # Load the saved dummy image.
     expected_size = 400
-    img = image_loading_utils.default_loader(filename, imsize=max_img_size,
-                                     bounding_box=[120, 120,
-                                                   120 + expected_size,
-                                                   120 + expected_size],
-                                     normalize=False)
+    img = image_loading_utils.default_loader(
+      filename, imsize=max_img_size,
+      bounding_box=[120, 120, 120 + expected_size, 120 + expected_size],
+      preprocess=False)
 
-    # Check the shape.
+    # Check that the final shape is as expected.
     self.assertAllEqual(tf.shape(img), [expected_size, expected_size, 3])
 
 

research/delf/delf/python/delg/DELG_INSTRUCTIONS.md

@@ -34,6 +34,8 @@ mv paris6k_images_tmp/paris/*/*.jpg paris6k_images/
 # Revisited annotations.
 wget http://cmp.felk.cvut.cz/revisitop/data/datasets/roxford5k/gnd_roxford5k.mat
 wget http://cmp.felk.cvut.cz/revisitop/data/datasets/rparis6k/gnd_rparis6k.mat
+wget http://cmp.felk.cvut.cz/cnnimageretrieval/data/test/roxford5k/gnd_roxford5k.pkl
+wget http://cmp.felk.cvut.cz/cnnimageretrieval/data/test/rparis6k/gnd_rparis6k.pkl
 ```
 
 ### Download model

research/delf/delf/python/detect_to_retrieve/__init__.py

@@ -20,5 +20,4 @@ from __future__ import print_function
 # pylint: disable=unused-import
 from delf.python.detect_to_retrieve import aggregation_extraction
 from delf.python.detect_to_retrieve import boxes_and_features_extraction
-from delf.python.detect_to_retrieve import dataset
 # pylint: enable=unused-import

research/delf/delf/python/pooling_layers/pooling.py

@@ -31,7 +31,6 @@ class MAC(tf.keras.layers.Layer):
     Args:
       x: [B, H, W, D] A float32 Tensor.
       axis: Dimensions to reduce. By default, dimensions [1, 2] are reduced.
-
     Returns:
       output: [B, D] A float32 Tensor.
     """
@@ -99,6 +98,47 @@ class GeM(tf.keras.layers.Layer):
     return gem(x, power=self.power, eps=self.eps, axis=axis)
 
 
+class GeMPooling2D(tf.keras.layers.Layer):
+
+  def __init__(self, power=20., pool_size=(2, 2), strides=None,
+               padding='valid', data_format='channels_last'):
+    """Generalized mean pooling (GeM) pooling operation for spatial data.
+
+    Args:
+      power: Float, power > 0. is an inverse exponent parameter (GeM power).
+      pool_size: Integer or tuple of 2 integers, factors by which to
+        downscale (vertical, horizontal)
+      strides: Integer, tuple of 2 integers, or None. Strides values.
+        If None, it will default to `pool_size`.
+      padding: One of `valid` or `same`. `valid` means no padding.
+        `same` results in padding evenly to the left/right or up/down of
+        the input such that output has the same height/width dimension as the
+        input.
+      data_format: A string, one of `channels_last` (default) or
+        `channels_first`. The ordering of the dimensions in the inputs.
+        `channels_last` corresponds to inputs with shape `(batch, height,
+        width, channels)` while `channels_first` corresponds to inputs with
+        shape `(batch, channels, height, width)`.
+    """
+    super(GeMPooling2D, self).__init__()
+    self.power = power
+    self.eps = 1e-6
+    self.pool_size = pool_size
+    self.strides = strides
+    self.padding = padding.upper()
+    data_format_conv = {'channels_last': 'NHWC',
+                        'channels_first': 'NCHW',
+                        }
+    self.data_format = data_format_conv[data_format]
+
+  def call(self, x):
+    tmp = tf.pow(x, self.power)
+    tmp = tf.nn.avg_pool(tmp, self.pool_size, self.strides,
+                         self.padding, self.data_format)
+    out = tf.pow(tmp, 1. / self.power)
+    return out
+
+
 def mac(x, axis=None):
   """Performs global max pooling (MAC).
 

research/delf/delf/python/pooling_layers/pooling_test.py

@@ -48,6 +48,37 @@ class PoolingsTest(tf.test.TestCase):
     # Compare actual and expected.
     self.assertAllClose(exp_output, result)
 
+  def testGeMPooling2D(self):
+    # Create a testing tensor.
+    x = tf.constant([[[1., 2., 3.],
+                      [4., 5., 6.],
+                      [7., 8., 9.]]])
+    x = tf.reshape(x, [1, 3, 3, 1])
+
+    # Checking GeMPooling2D relation to MaxPooling2D for the large values of
+    # `p`.
+    max_pool_2d = tf.keras.layers.MaxPooling2D(pool_size=(2, 2),
+                                               strides=(1, 1), padding='valid')
+    out_max = max_pool_2d(x)
+    gem_pool_2d = pooling.GeMPooling2D(power=30., pool_size=(2, 2),
+                                       strides=(1, 1), padding='valid')
+    out_gem_max = gem_pool_2d(x)
+
+    # Check that for large `p` GeMPooling2D is close to MaxPooling2D.
+    self.assertAllEqual(out_max, tf.round(out_gem_max))
+
+    # Checking GeMPooling2D relation to AveragePooling2D for the value
+    # of `p` = 1.
+    avg_pool_2d = tf.keras.layers.AveragePooling2D(pool_size=(2, 2),
+                                                   strides=(1, 1),
+                                                   padding='valid')
+    out_avg = avg_pool_2d(x)
+    gem_pool_2d = pooling.GeMPooling2D(power=1., pool_size=(2, 2),
+                                       strides=(1, 1), padding='valid')
+    out_gem_avg = gem_pool_2d(x)
+    # Check that for `p` equals 1., GeMPooling2D becomes AveragePooling2D.
+    self.assertAllEqual(out_avg, out_gem_avg)
+
 
 if __name__ == '__main__':
   tf.test.main()