Commit fbf76072 authored by jay's avatar jay
Browse files

Incorporates dynamic func loadings for optional CUDA support

parent 69cf4879

autocnet/__init__.py

+37 −8
Changes for autocnet/__init__.py: 37 added lines, 8 removed lines.
Original line number Diff line number Diff line 
import os

import autocnet



__version__ = "0.1.0"



def get_data(filename):

    packagdir = autocnet.__path__[0]

    dirname = os.path.join(os.path.dirname(packagdir), 'data')

    fullname = os.path.join(dirname, filename)

    return fullname



import autocnet.examples

import autocnet.camera

import autocnet.cg
@@ -18,3 +10,40 @@ import autocnet.matcher 
import autocnet.transformation

import autocnet.utils

import autocnet.utils



__version__ = "0.1.0"



def get_data(filename):

    packagdir = autocnet.__path__[0]

    dirname = os.path.join(os.path.dirname(packagdir), 'data')

    fullname = os.path.join(dirname, filename)

    return fullname



def cuda(enable=False, gpu=0):

    # Classes/Methods that can vary if GPU is available

    from autocnet.graph.node import Node

    from autocnet.graph.edge import Edge

    if enable:

        print('Enabling CUDA')

        try:

            import cudasift as cs

            cs.PyInitCuda(gpu)



            # Here is where the GPU methods get patched into the class

            from autocnet.matcher.cuda_extractor import extract_features 

            Node._extract_features = staticmethod(extract_features)



            from autocnet.matcher.cuda_matcher import match

            Edge.match = match

        except Exception:

            print('Failed to enable cuda')

        return



    print('CUDA Disabled')

    # Here is where the CPU methods get patched into the class

    from autocnet.matcher.feature_extractor import extract_features

    Node._extract_features = staticmethod(extract_features)



    from autocnet.matcher.feature_matcher import match

    Edge.match = match

cuda()

autocnet/graph/edge.py

+33 −40
Changes for autocnet/graph/edge.py: 33 added lines, 40 removed lines.
Original line number Diff line number Diff line
@@ -5,6 +5,7 @@ import numpy as np 
import pandas as pd

from scipy.spatial.distance import cdist



import autocnet

from autocnet.utils import utils

from autocnet.matcher import health

from autocnet.matcher import outlier_detector as od
@@ -77,9 +78,13 @@ class Edge(dict, MutableMapping): 
        # state, dynamically draw the mask from the object.

        for c in self._masks.columns:

            if c in mask_lookup:

                try:

                    truncated_mask = getattr(self, mask_lookup[c]).mask

                    self._masks[c] = False

                    self._masks[c].iloc[truncated_mask.index] = truncated_mask

                except Exception:

                    #TODO: Get rid of state

                    pass

        return self._masks



    @masks.setter
@@ -342,48 +347,36 @@ class Edge(dict, MutableMapping): 
        k : int

            The number of neighbors to find

        """

        def mono_matches(a, b, aidx=None, bidx=None):

            """

            Apply the FLANN match_features



            Parameters

            ----------

            a : object

                A node object



            b : object

                A node object



            aidx : iterable

                   An index for the descriptors to subset

        pass



            bidx : iterable

                   An index for the descriptors to subset

    def cuda_match(self, ratio=0.8, **kwargs):

        """

            # Subset if requested

            if aidx is not None:

                ad = a.descriptors[aidx]

            else:

                ad = a.descriptors



            if bidx is not None:

                bd = b.descriptors[bidx]

            else:

                bd = b.descriptors



            # Load, train, and match

            fl.add(ad, a.node_id, index=aidx)

            fl.train()

            matches = fl.query(bd, b.node_id, k, index=bidx)

            self._add_matches(matches)

            fl.clear()



        fl = FlannMatcher()

        mono_matches(self.source, self.destination)

        mono_matches(self.destination, self.source)







        Apply a composite CUDA matcher and ratio check.  If this method is used,

        no additional ratio check is necessary and no symmetry check is required.

        The ratio check is embedded on the cuda side and returned as an

        ambiguity value.  In testing symmetry is not required as it is expensive

        without significant gain in accuracy when using this implementation.

        """

        if not autocnet.cudasift:

            warnings.warn('CudaSift is not available, please use the standard matcher.')

        s_siftdata = autocnet.cs.PySiftData.from_data_frame(self.source.get_keypoints(), self.source.descriptors)

        d_siftdata = autocnet.cs.PySiftData.from_data_frame(self.destination.get_keypoints(), self.destination.descriptors)



        autocnet.cs.PyMatchSiftData(s_siftdata, d_siftdata)

        matches, _ = s_siftdata.to_data_frame()

        source = np.empty(len(matches))

        source[:] = self.source.node_id

        destination = np.empty(len(matches))

        destination[:] = self.destination.node_id





        df = pd.concat([pd.Series(source), pd.Series(matches.index),

                        pd.Series(destination), matches.match,

                        matches.score, matches.ambiguity], axis=1)

        df.columns = ['source_image', 'source_idx', 'destination_image',

                        'destination_idx', 'score', 'ambiguity']

        print(df)

        self.matches = df

    def _add_matches(self, matches):

        """

        Given a dataframe of matches, either append to an existing

autocnet/graph/network.py

+5 −3
Changes for autocnet/graph/network.py: 5 added lines, 3 removed lines.
Original line number Diff line number Diff line
@@ -206,7 +206,7 @@ class CandidateGraph(nx.Graph): 


        raise NotImplementedError



    def extract_features(self, method='orb', extractor_parameters={}):

    def extract_features(self, *args, **kwargs):

        """

        Extracts features from each image in the graph and uses the result to assign the

        node attributes for 'handle', 'image', 'keypoints', and 'descriptors'.
@@ -224,8 +224,7 @@ class CandidateGraph(nx.Graph): 
        """

        for i, node in self.nodes_iter(data=True):

            image = node.get_array()

            node.extract_features(image, method=method,

                                  extractor_parameters=extractor_parameters)

            node.extract_features(image, *args, **kwargs),



    def save_features(self, out_path, nodes=[]):

        """
@@ -295,6 +294,9 @@ class CandidateGraph(nx.Graph): 
        """

        self.apply_func_to_edges('match', *args, **kwargs)



    def cuda_match(self, *args, **kwargs):

        self.apply_func_to_edges('cuda_match', *args, **kwargs)



    def decompose_and_match_features(self, *args, **kwargs):

        """

        For all edges in the graph, apply coupled decomposition followed by

autocnet/graph/node.py

+6 −2
Changes for autocnet/graph/node.py: 6 added lines, 2 removed lines.
Original line number Diff line number Diff line
@@ -227,7 +227,8 @@ class Node(dict, MutableMapping): 


        return keypoints



    def extract_features(self, array, **kwargs):

    @staticmethod

    def _extract_features(*args, **kwargs):

        """

        Extract features for the node
@@ -239,7 +240,10 @@ class Node(dict, MutableMapping): 
                 kwargs passed to autocnet.feature_extractor.extract_features



        """

        self._keypoints, self.descriptors = fe.extract_features(array, **kwargs)

        pass



    def extract_features(self, *args, **kwargs):

        self._keypoints, self.descriptors = Node._extract_features(*args, **kwargs)



    def load_features(self, in_path):

        """

autocnet/matcher/cuda_extractor.py

0 → 100644
+19 −0
Changes for autocnet/matcher/cuda_extractor.py: 19 added lines, 0 removed lines.
Original line number Diff line number Diff line 
import warnings



import cudasift as cs



def extract_features(array, nfeatures=None): 

    if not nfeatures:

        nfeatures = int(max(array.shape) / 1.75)

    else:

        warnings.warn('NFeatures specified with the CudaSift implementation.  Please ensure the distribution of keypoints is what you expect.')

    

    siftdata = cs.PySiftData(nfeatures)

    cs.ExtractKeypoints(array, siftdata)

    keypoints, descriptors = siftdata.to_data_frame()

    keypoints = keypoints[['xpos', 'ypos', 'scale', 'sharpness', 'edgeness', 'orientation', 'score', 'ambiguity']]

    # Set the columns that have unfilled values to zero to avoid confusion

    keypoints['score'] = 0.0

    keypoints['ambiguity'] = 0.0



    return keypoints, descriptors