code/libs/utils/Evaluation/utils.py

import json
#import urllib2

import numpy as np
import contextlib
import joblib
from tqdm import tqdm


API = 'http://ec2-52-11-11-89.us-west-2.compute.amazonaws.com/challenge16/api.py'

@contextlib.contextmanager
def tqdm_joblib(tqdm_object):
    """Context manager to patch joblib to report into tqdm progress bar given as argument"""
    class TqdmBatchCompletionCallback(joblib.parallel.BatchCompletionCallBack):
        def __call__(self, *args, **kwargs):
            tqdm_object.update(n=self.batch_size)
            return super().__call__(*args, **kwargs)

    old_batch_callback = joblib.parallel.BatchCompletionCallBack
    joblib.parallel.BatchCompletionCallBack = TqdmBatchCompletionCallback
    try:
        yield tqdm_object
    finally:
        joblib.parallel.BatchCompletionCallBack = old_batch_callback
        tqdm_object.close()

def get_blocked_videos(api=API):
    api_url = '{}?action=get_blocked'.format(api)
    req = urllib2.Request(api_url)
    response = urllib2.urlopen(req)
    return json.loads(response.read())

def interpolated_prec_rec(prec, rec):
    """Interpolated AP - VOCdevkit from VOC 2011.
    """
    mprec = np.hstack([[0], prec, [0]])
    mrec = np.hstack([[0], rec, [1]])
    for i in range(len(mprec) - 1)[::-1]:
        mprec[i] = max(mprec[i], mprec[i + 1])
    idx = np.where(mrec[1::] != mrec[0:-1])[0] + 1
    ap = np.sum((mrec[idx] - mrec[idx - 1]) * mprec[idx])
    return ap

def segment_iou(target_segment, candidate_segments):
    """Compute the temporal intersection over union between a
    target segment and all the test segments.

    Parameters
    ----------
    target_segment : 1d array
        Temporal target segment containing [starting, ending] times.
    candidate_segments : 2d array
        Temporal candidate segments containing N x [starting, ending] times.

    Outputs
    -------
    tiou : 1d array
        Temporal intersection over union score of the N's candidate segments.
    """
    tt1 = np.maximum(target_segment[0], candidate_segments[:, 0])
    tt2 = np.minimum(target_segment[1], candidate_segments[:, 1])
    # Intersection including Non-negative overlap score.
    segments_intersection = (tt2 - tt1).clip(0)
    # Segment union.
    segments_union = (candidate_segments[:, 1] - candidate_segments[:, 0]) \
      + (target_segment[1] - target_segment[0]) - segments_intersection
    # Compute overlap as the ratio of the intersection
    # over union of two segments.
    tIoU = segments_intersection.astype(float) / segments_union
    return tIoU

def wrapper_segment_iou(target_segments, candidate_segments):
    """Compute intersection over union btw segments
    Parameters
    ----------
    target_segments : ndarray
        2-dim array in format [m x 2:=[init, end]]
    candidate_segments : ndarray
        2-dim array in format [n x 2:=[init, end]]
    Outputs
    -------
    tiou : ndarray
        2-dim array [n x m] with IOU ratio.
    Note: It assumes that candidate-segments are more scarce that target-segments
    """
    if candidate_segments.ndim != 2 or target_segments.ndim != 2:
        raise ValueError('Dimension of arguments is incorrect')

    n, m = candidate_segments.shape[0], target_segments.shape[0]
    tiou = np.empty((n, m))
    for i in xrange(m):
        tiou[:, i] = segment_iou(target_segments[i,:], candidate_segments)

    return tiou