def main(dataset='measurements.npy'):
data = np.load(dataset, mmap_mode='r')
print 'loaded', dataset, data.shape
fig = plt.figure()
ax = util.axes(fig, 111)
trial = data[15, 1, 5]
for f in range(0, len(trial), 300):
util.plot_skeleton(ax, trial[f], alpha=1)
x, y, z = trial[:, TARGET].T
ax.plot(x, z, y, 'o-', color='#111111', alpha=0.5)
util.set_limits(ax, center=(0, 0, 1), span=1)
ax.w_xaxis.set_pane_color((1, 1, 1, 1))
ax.w_yaxis.set_pane_color((1, 1, 1, 1))
ax.w_zaxis.set_pane_color((1, 1, 1, 1))
#plt.gcf().set_size_inches(12, 10)
#plt.savefig('single-trial.pdf', dpi=600)
plt.show()
return transformed_frames[:, :, :2]
# apply T
transformed_sequences = np.asarray(
[_transform_sequence(seq) for seq in sequences])
return transformed_sequences
if __name__ == '__main__':
# read in sequence
dataset_dir = Path(__file__).parent / '../../datasets/KTH_Action_Dataset/'
metadata_file = dataset_dir / 'metadata.csv'
from datasets import KTHDataset
dataset = KTHDataset(metadata_file,
dataset_dir,
use_confidence_scores=False)
seqs, actions = dataset[:4]
original_seq = np.copy(seqs[0])
augmented_sequences = augment_data(seqs[:4])
assert np.array_equal(seqs[0],
original_seq) # check original sequence still intact
from util import plot_skeleton
print('Saving plot of skeleton')
plot_skeleton(augmented_sequences[0],
'../../datasets/example_augmented_plot.mp4')
def main(dataset='measurements.npy'):
data = np.load(dataset, mmap_mode='r')
print 'loaded', dataset, data.shape
plots = list(range(N * N))
frames = [[] for _ in plots]
for subj in data:
for block in subj[1:]:
for trial in block:
if trial[0, C.col('trial-hand')] == C.right:
for frame in trial:
for i in plots:
if within_region(frame, i):
frames[i].append(frame)
break
u, v = np.mgrid[0:2 * np.pi:11j, 0:np.pi:7j]
sphx = np.cos(u) * np.sin(v)
sphy = np.sin(u) * np.sin(v)
sphz = np.cos(v)
fig = plt.figure()
for i, postures in enumerate(frames):
if not postures:
continue
if i != 2:
continue
postures = np.array(postures)
for m in range(50):
marker = postures[:, 17+m*4:17+(m+1)*4]
drops = marker[:, 3] < 0
marker[drops, :3] = marker[~drops, :3].mean(axis=0)
means = postures.mean(axis=0)
stds = postures.std(axis=0)
#ax = util.axes(fig, 111)
#for frame in postures[::5]:
# util.plot_skeleton(ax, frame, alpha=0.1)
ax = util.axes(fig, 110 * N + i + 1)
util.plot_skeleton(ax, means, alpha=1.0)
for m in range(50):
mx, my, mz = means[17+m*4:20+m*4]
sx, sy, sz = stds[17+m*4:20+m*4] / 2
ax.plot_wireframe(sphx * sx + mx, sphz * sz + mz, sphy * sy + my,
color=C.MARKER_COLORS[m], alpha=0.3)
#tgtx, tgty, tgtz = postures.mean(axis=0)[
# C.cols('target-x', 'target-y', 'target-z')]
#ax.plot([tgtx], [tgtz], [tgty], 'o', color='#111111')
#for m in range(50):
# marker = postures[:, 17 + 4 * m:17 + 4 * (m+1)]
# position = marker.mean(axis=0)
# size = marker.std(axis=0)
# ax.plot_surface()
util.set_limits(ax, center=(0, -0.5, 1), span=1)
ax.w_xaxis.set_pane_color((1, 1, 1, 1))
ax.w_yaxis.set_pane_color((1, 1, 1, 1))
ax.w_zaxis.set_pane_color((1, 1, 1, 1))
ax.set_title(['Top Right', 'Top Left', 'Bottom Right', 'Bottom Left'][i])
#for m in range(50):
# x, z, y = frame[m*4:m*4+3]
# ax.text(x, y, z, str(m))
plt.gcf().set_size_inches(12, 10)
#plt.savefig('reach-targets-with-variance.pdf', dpi=600)
plt.show()
