added faster_rcnn model; py3 compatible

general/conv_calc.py

@@ -0,0 +1,73 @@
+#! /usr/bin/env python
+
+# Copyright (c) 2018 Uber Technologies, Inc.
+# 
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the "Software"), to deal
+# in the Software without restriction, including without limitation the rights
+# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+# copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in all
+# copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+# SOFTWARE.
+
+
+# Functions for calculating convnet receptive field sizes
+# Can use, e.g., like this:
+# rso = [(1,1,0)]                  Input layer receptive field size (1 pixel), stride (1 pixel), and offset (0 pixels)
+# rso.append(receptive_calc_conv(5, 1)(rso[-1]))
+# rso.append(receptive_calc_pool(2, 2)(rso[-1]))
+# rso.append(receptive_calc_conv_same(5, 1)(rso[-1]))
+# rso.append(receptive_calc_conv_same(5, 1)(rso[-1]))
+# print rso
+# [(1, 1, 0), (5, 1, 0), (6, 2, 0), (14, 2, -4), (22, 2, -8)]
+# print rso[-1]
+# (22, 2, -8)                      Output layer receptive field size (22 pixels), stride (2 pixels), and offset (-8 pixels)
+
+import numpy as np
+
+
+def receptive_calc_conv(filt_hw, filt_stride):
+    '''Return a function from rso -> rso. rso is a tuple of (receptive field hw, stride, and offset). Assumes 'valid' convolution.'''
+    def fn(rso):
+        rec_hw, rec_stride, rec_offset = rso
+        return (
+            rec_hw + rec_stride*(filt_hw-1),
+            rec_stride * filt_stride,
+            rec_offset
+        )
+    return fn
+
+# Same function works for pool
+receptive_calc_pool = receptive_calc_conv
+
+def receptive_calc_conv_same(filt_hw, filt_stride):
+    '''Return a function from rso -> rso. rso is a tuple of (receptive field hw, stride, and offset). Assumes 'same' convolution.'''
+    assert np.mod(filt_hw, 2) == 1, 'receptive_calc_conv_same only works for convolution with odd-sized filters'
+    def fn(rso):
+        rec_hw, rec_stride, rec_offset = rso
+        return (
+            rec_hw + rec_stride*(filt_hw-1),
+            rec_stride * filt_stride,
+            rec_offset - (filt_hw-1)/2 * rec_stride
+        )
+    return fn
+
+def receptive_calc_pad(pad_hw):
+    def fn(rso):
+        rec_hw, rec_stride, rec_offset = rso
+        return (
+            rec_hw,
+            rec_stride,
+            rec_offset - pad_hw*rec_stride,
+        )
+    return fn

general/lr_policy.py

@@ -1,3 +1,25 @@
+
+# Copyright (c) 2018 Uber Technologies, Inc.
+# 
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the "Software"), to deal
+# in the Software without restriction, including without limitation the rights
+# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+# copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in all
+# copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+# SOFTWARE.
+
+
 import numpy as np
 from scipy import stats
 from IPython import embed
@@ -44,7 +66,7 @@ def get_lr(self, buddy):
             return None
         if self.n_steps < self.max_steps and buddy.epoch - self.last_step_at >= self.step_every:
             new_lr = self.lr * self.step_ratio
-            print 'Changing LR from %g to %g' % (self.lr, new_lr)
+            print('Changing LR from %g to %g' % (self.lr, new_lr))
             self.lr = new_lr
             self.n_steps += 1
             self.last_step_at = buddy.epoch
@@ -97,7 +119,7 @@ def get_lr(self, buddy):
 
         if self.n_steps < self.max_steps and buddy.epoch - self.last_step_at >= self.step_every:
             new_lr = self.lr * self.step_ratio
-            print 'Changing LR from %g to %g' % (self.lr, new_lr)
+            print('Changing LR from %g to %g' % (self.lr, new_lr))
             self.lr = new_lr
             self.n_steps += 1
             self.last_step_at = buddy.epoch

general/tfutil.py

@@ -1,3 +1,24 @@
+
+# Copyright (c) 2018 Uber Technologies, Inc.
+# 
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the "Software"), to deal
+# in the Software without restriction, including without limitation the rights
+# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+# copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in all
+# copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+# SOFTWARE.
+
 import os
 import numpy as np
 from orderedset import OrderedSet
@@ -406,7 +427,7 @@ def log_scalars(writer, iters, scalars_dict, prefix=None):
     if len(prefix) > 0 and not prefix.endswith('/'):
         prefix = prefix + '/'
 
-    for key, val in scalars_dict.iteritems():
+    for key, val in scalars_dict.items():
         if hasattr(val, 'dtype'):
             val = np.asscalar(val)   # Convert, e.g., numpy.float32 -> float
         writer.add_summary(tf.Summary(value=[tf.Summary.Value(tag='%s%s' % (prefix, key), simple_value=val)]).SerializeToString(), iters)
@@ -472,19 +493,19 @@ def add_grads_and_vars_hist_summaries(grads_and_vars):
 def add_grad_summaries(grads_and_vars, add_summaries_train=True, quiet=False):
     # Add summary nodes for grad values and prints a summary as well
     if not quiet:
-        print '\nGrads:'
+        print('\nGrads:')
         if len(grads_and_vars) == 0:
-            print '  <None>'
+            print('  <None>')
     for grad, var in grads_and_vars:
         if grad is None:
             continue   # skip grads that are None (corner case: not computed because model.loss has no dependence?)
         grad_name = '%s/%s__grad' % tuple(var.name.rsplit('/', 1))
         if not quiet:
-            print '  ', grad_name, grad
+            print('  ', grad_name, grad)
         if add_summaries_train:
             hist_summaries_train(grad, name=grad_name)
     if not quiet:
-        print
+        print()
 
 
 ########################
@@ -547,8 +568,8 @@ def sess_run_dict(sess, fetch_names, fetch_vars=None, feed_dict=None, options=No
     if dict_mode:
         assert fetch_vars is None, 'provide either dict or list of names and vars, not both'
         fetch_dict = fetch_names
-        fetch_names = fetch_dict.keys()
-        fetch_vars = fetch_dict.values()
+        fetch_names = list(fetch_dict.keys())
+        fetch_vars = list(fetch_dict.values())
 
     assert len(fetch_names) == len(fetch_vars), 'length of fetch_names must match length of fetch_vars'
     assert isinstance(fetch_vars, list) or isinstance(fetch_vars, tuple), 'fetch_vars should be list or tuple'
@@ -588,15 +609,15 @@ def summarize_weights(weights, sess=None):
     titl = '  %50s: %10s %-20s' % ('NAME', 'SIZE', 'SHAPE')
     if sess:
         titl += ' %10s, %10s, %10s' % ('MIN', 'MAX', 'RMS')
-    print titl
+    print(titl)
     for ii,var in enumerate(weights):
         st = '  %50s: %10d %-20s' % (var.name, np.prod(var.get_shape().as_list()), var.get_shape().as_list())
         if sess:
             val = vals[ii]
             st += ' %10s, %10s, %10s' % ('%.3g' % val.min(), '%.3g' % val.max(), '%.3g' % np.sqrt((val**2).mean()))
-        print st
+        print(st)
         total_params += np.prod(var.get_shape().as_list())
-    print '  %50s: %10d' % ('Total', total_params)
+    print('  %50s: %10d' % ('Total', total_params))
     return total_params
 
 
@@ -605,16 +626,16 @@ def val_or_dynamic(vord):
 
 
 def summarize_opt(opt):
-    print 'Optimizer:'
-    print '   ', opt
+    print('Optimizer:')
+    print('   ', opt)
     if isinstance(opt, tf.train.MomentumOptimizer):
-        print '   LR: %s, momentum: %g, use_nesterov: %s' % (val_or_dynamic(opt._learning_rate), opt._momentum, opt._use_nesterov)
+        print('   LR: %s, momentum: %g, use_nesterov: %s' % (val_or_dynamic(opt._learning_rate), opt._momentum, opt._use_nesterov))
     elif isinstance(opt, tf.train.RMSPropOptimizer):
-        print '   LR: %s, momentum: %g, decay: %g, epsilon: %g' % (val_or_dynamic(opt._learning_rate), opt._momentum, opt._decay, opt._epsilon)
+        print('   LR: %s, momentum: %g, decay: %g, epsilon: %g' % (val_or_dynamic(opt._learning_rate), opt._momentum, opt._decay, opt._epsilon))
     elif isinstance(opt, tf.train.AdamOptimizer):
-        print '   LR: %s, beta1: %g, beta2: %g, epsilon: %g' % (val_or_dynamic(opt._lr), opt._beta1, opt._beta2, opt._epsilon)
+        print('   LR: %s, beta1: %g, beta2: %g, epsilon: %g' % (val_or_dynamic(opt._lr), opt._beta1, opt._beta2, opt._epsilon))
     else:
-        print '   (cannot summarize unknown type of optimizer)'
+        print('   (cannot summarize unknown type of optimizer)')
 
 
 def tf_assert_gpu(sess):
@@ -624,8 +645,8 @@ def tf_assert_gpu(sess):
     try:
         sess.run(bar, {foo: 1})
     except:
-        print '\n\n\ntf_assert_gpu: no GPU is present! In case it helps, CUDA_VISIBLE_DEVICES is %s' % repr(os.environ.get('CUDA_VISIBLE_DEVICES', None))
-        print 'See error below:\n\n\n'
+        print('\n\n\ntf_assert_gpu: no GPU is present! In case it helps, CUDA_VISIBLE_DEVICES is %s' % repr(os.environ.get('CUDA_VISIBLE_DEVICES', None)))
+        print('See error below:\n\n\n')
         raise
 
 

general/util.py

@@ -1,3 +1,24 @@
+
+# Copyright (c) 2018 Uber Technologies, Inc.
+# 
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the "Software"), to deal
+# in the Software without restriction, including without limitation the rights
+# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+# copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in all
+# copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+# SOFTWARE.
+
 import os
 import errno
 import time
@@ -8,11 +29,11 @@ class DuckStruct(object):
     '''Use to store anything!'''
 
     def __init__(self, **kwargs):
-        for k,v in kwargs.items():
+        for k,v in list(kwargs.items()):
             setattr(self, k, v)
 
     def __repr__(self):
-        rep = ['%s=%s' % (k, repr(v)) for k,v in self.__dict__.items()]
+        rep = ['%s=%s' % (k, repr(v)) for k,v in list(self.__dict__.items())]
         return 'DuckStruct(%s)' % ', '.join(rep)
 
 
@@ -26,10 +47,10 @@ def __init__(self, *args, **kwargs):
         super(DotDict, self).__init__(*args, **kwargs)
         for arg in args:
             if isinstance(arg, dict):
-                for k, v in arg.iteritems():
+                for k, v in arg.items():
                     self[k] = v
         if kwargs:
-            for k, v in kwargs.iteritems():
+            for k, v in kwargs.items():
                 self[k] = v
 
     def __getattr__(self, attr):
@@ -91,7 +112,7 @@ def __enter__(self):
     def __exit__(self, *args):
         if not self.quiet:
             titlestr = (' ' + self.title) if self.title else ''
-            print 'Elapsed%s: wall: %.06f, sys: %.06f' % ((titlestr,) + self.elapsed())
+            print('Elapsed%s: wall: %.06f, sys: %.06f' % ((titlestr,) + self.elapsed()))
 
 
 class TicToc(object):