#!/usr/bin/env python2.7
# Li Ding
# Mar. 2018
# Modified from https://github.com/colincsl/TemporalConvolutionalNetworks/blob/master/code/utils.py
from __future__ import division
import matplotlib.pylab as plt
import numpy as np
import os
import scipy.io as sio
import cv2
import glob
from tqdm import tqdm
plt.rcParams['figure.figsize'] = [10., 10.]
# ------------- Visualization -------------
def imshow_(x, **kwargs):
if x.ndim == 2:
plt.imshow(x, interpolation="nearest", **kwargs)
elif x.ndim == 1:
plt.imshow(x[:, None].T, interpolation="nearest", **kwargs)
plt.yticks([])
plt.axis("tight")
# ------------- Data -------------
def mask_data(X, Y, max_len=None, mask_value=0):
if max_len is None:
max_len = np.max([x.shape[0] for x in X])
X_ = np.zeros([len(X), max_len, X[0].shape[1]]) + mask_value
Y_ = np.zeros([len(X), max_len, Y[0].shape[1]]) + mask_value
mask = np.zeros([len(X), max_len])
for i in range(len(X)):
l = X[i].shape[0]
X_[i, :l] = X[i]
Y_[i, :l] = Y[i]
mask[i, :l] = 1
return X_, Y_, mask[:, :, None]
# Unmask data
def unmask(X, M):
if X[0].ndim == 1 or (X[0].shape[0] > X[0].shape[1]):
return [X[i][M[i].flatten() > 0] for i in range(len(X))]
else:
return [X[i][:, M[i].flatten() > 0] for i in range(len(X))]
def match_lengths(X, Y, n_feat):
# Check lengths of data and labels match
if X[0].ndim == 1 or (X[0].shape[0] == n_feat):
for i in range(len(Y)):
length = min(X[i].shape[1], Y[i].shape[0])
X[i] = X[i][:, :length]
Y[i] = Y[i][:length]
else:
for i in range(len(Y)):
length = min(X[i].shape[0], Y[i].shape[0])
X[i] = X[i][:length]
Y[i] = Y[i][:length]
return X, Y
def remap_labels(Y_all):
# Map arbitrary set of labels (e.g. {1,3,5}) to contiguous sequence (e.g. {0,1,2})
ys = np.unique([np.hstack([np.unique(Y_all[i]) for i in range(len(Y_all))])])
y_max = ys.max()
y_map = np.zeros(y_max + 1, np.int) - 1
for i, yi in enumerate(ys):
y_map[yi] = i
Y_all = [y_map[Y_all[i]] for i in range(len(Y_all))]
return Y_all
def max_seg_count(Y):
def seg_count(y):
# Input label sequence
return len(segment_labels(y))
# Input: list of label sequences
return max(map(seg_count, Y))
def subsample(X, Y, rate=1, dim=0):
if dim == 0:
X_ = [x[::rate] for x in X]
Y_ = [y[::rate] for y in Y]
elif dim == 1:
X_ = [x[:, ::rate] for x in X]
Y_ = [y[::rate] for y in Y]
else:
print("Subsample not defined for dim={}".format(dim))
return None, None
return X_, Y_
# ------------- Segment functions -------------
def segment_labels(Yi):
idxs = [0] + (np.nonzero(np.diff(Yi))[0] + 1).tolist() + [len(Yi)]
Yi_split = np.array([Yi[idxs[i]] for i in range(len(idxs) - 1)])
return Yi_split
def segment_data(Xi, Yi):
idxs = [0] + (np.nonzero(np.diff(Yi))[0] + 1).tolist() + [len(Yi)]
Xi_split = [np.squeeze(Xi[:, idxs[i]:idxs[i + 1]]) for i in range(len(idxs) - 1)]
Yi_split = np.array([Yi[idxs[i]] for i in range(len(idxs) - 1)])
return Xi_split, Yi_split
def segment_intervals(Yi):
idxs = [0] + (np.nonzero(np.diff(Yi))[0] + 1).tolist() + [len(Yi)]
intervals = [(idxs[i], idxs[i + 1]) for i in range(len(idxs) - 1)]
return intervals
def segment_lengths(Yi):
idxs = [0] + (np.nonzero(np.diff(Yi))[0] + 1).tolist() + [len(Yi)]
intervals = [(idxs[i + 1] - idxs[i]) for i in range(len(idxs) - 1)]
return np.array(intervals)
# ------------- IO -------------
def save_predictions(dir_out, y_pred, y_truth, idx_task, experiment_name=""):
if experiment_name != "":
dir_out += "/{}/".format(experiment_name)
# Make sure fiolder exists
if not os.path.isdir(dir_out):
os.makedirs(dir_out)
truth_test_all_out = {"t{}_{}".format(idx_task, k): v for (k, v) in enumerate(y_truth)}
predict_test_all_out = {"t{}_{}".format(idx_task, k): v for k, v in enumerate(y_pred)}
sio.savemat(dir_out + "/{}_truth.mat".format(idx_task), truth_test_all_out)
sio.savemat(dir_out + "/{}_predict.mat".format(idx_task), predict_test_all_out)
# ------------- Vision -------------
def load_images(uris, rez_im, uri_data):
# Load images for CNN
X = np.empty((len(uris), 3, rez_im, rez_im), dtype=np.float32)
for i, x in enumerate(uris):
im = cv2.imread(uri_data + x)
im = cv2.resize(im, (rez_im, rez_im))
X[i] = im.T
return X
def check_images_available(x_uri, y, uri_data):
# Check if there are any missing files
no_file = []
for i, x in enumerate(x_uri):
if not os.path.isfile(uri_data + x):
# print("Missing", x)
no_file += [i]
x_uri = np.array([x_uri[i] for i in range(len(x_uri)) if i not in no_file])
y = np.array([y[i] for i in range(len(y)) if i not in no_file])
if len(no_file) > 0:
print("Missing #", len(no_file))
return x_uri, y
# ----------- dataset -------------
class breakfast_dataset():
def __init__(self, dir):
print 'loading data from:', dir
all = ["P%02d" % i for i in range(3, 56)]
splits = [all[:13], all[13:26], all[26:39], all[39:52]]
features = glob.glob(os.path.join(dir, 'breakfast_data', 's1', '*', '*.txt'))
labels = glob.glob(os.path.join(dir, 'segmentation_coarse', '*', '*.txt'))
features.sort()
labels.sort()
self.data = [[] for i in range(8)] # x,y for 4 splits
actions = []
for f, l in tqdm(zip(features, labels)):
_x = np.loadtxt(f)
_y_raw = open(l).readlines()
_y = np.repeat(0, int(_y_raw[-1].split()[0].split('-')[1]))
for act in _y_raw:
tm, lb = act.split()
if lb not in actions:
actions.append(lb)
_y[(int(tm.split('-')[0]) - 1):int(tm.split('-')[1])] = actions.index(lb)
n = min(len(_x), len(_y)) # make sure the same length
for i in range(4):
if sum([k in f for k in splits[i]]) > 0: # in the split
self.data[2 * i].append(_x[5:n, 1:]) # exclude index and first 5 frames (all-zero feature)
self.data[2 * i + 1].append(_y[5:n])
break
def get_split(self, split, type):
if split not in ['s1', 's2', 's3', 's4']:
print "please enter split index from ['s1','s2','s3','s4']"
else:
split = int(split[1]) - 1
if type == 'train':
x = []
y = []
for i in range(4):
if i != split:
x += self.data[2 * i]
y += self.data[2 * i + 1]
return x, y
elif type == 'test':
return self.data[2 * split], self.data[2 * split + 1]
else:
print "please enter type from ['train','test']"
