def compute_responses(mlps_dir):
stim_type = 'breakfast'
spatial_downsample_factor = 4
if stim_type == 'breakfast':
video_dir = './temp/breakfast_sorted/cam01/'
features_dir = os.path.join(
'./temp/video_features/',
os.path.basename(os.path.normpath(mlps_dir)))
else:
raise NotImplementedError
print 'Reading in the videos from %s...' % mlps_dir
videos = read_videos(video_dir,
spatial_downsample_factor=spatial_downsample_factor)
print 'Read.'
action_names = os.listdir(video_dir)
print 'Classes : ', action_names
for i in range(11):
print 'Using MLP %d...' % i
with open(os.path.join(mlps_dir, 'mlp_%d' % i), 'rb') as f:
Wh, bh, Wo, bo = pickle.load(f)
mlp = create_mlp_from_params(Wh, bh, Wo, bo, rectify=True)
print 'MLP created from saved parameters. Computing MLP output...'
sim_rsps = [[transform(mlp, video) for video in class_videos]
for class_videos in videos]
print 'Storing MLP output on %s videos...' % stim_type
mlp_output_dir = os.path.join(features_dir, 'mlp_%d' % i)
dump_features(mlp_output_dir, sim_rsps, action_names)
print 'Stored.\n'
def compute_responses(mlps_dir):
stim_type = 'breakfast'
spatial_downsample_factor = 4
if stim_type == 'breakfast':
video_dir = './temp/breakfast_sorted/cam01/'
features_dir = os.path.join('./temp/video_features/',
os.path.basename(
os.path.normpath(mlps_dir)))
else:
raise NotImplementedError
print 'Reading in the videos from %s...' % mlps_dir
videos = read_videos(video_dir,
spatial_downsample_factor=spatial_downsample_factor)
print 'Read.'
action_names = os.listdir(video_dir)
print 'Classes : ', action_names
for i in range(11):
print 'Using MLP %d...' % i
with open(os.path.join(mlps_dir, 'mlp_%d' % i), 'rb') as f:
Wh, bh, Wo, bo = pickle.load(f)
mlp = create_mlp_from_params(Wh, bh, Wo, bo, rectify=True)
print 'MLP created from saved parameters. Computing MLP output...'
sim_rsps = [[transform(mlp, video) for video in class_videos]
for class_videos in videos]
print 'Storing MLP output on %s videos...' % stim_type
mlp_output_dir = os.path.join(features_dir, 'mlp_%d' % i)
dump_features(mlp_output_dir, sim_rsps, action_names)
print 'Stored.\n'
def get_simulated_responses(model_dir, videos):
responses = []
for i in range(11):
with open(os.path.join(model_dir, 'mlp_%d' % i), 'rb') as f:
Wh, bh, Wo, bo = pickle.load(f)
mlp = create_mlp_from_params(Wh, bh, Wo, bo, rectify=True)
sim_rsps = [[transform(mlp, video) for video in class_videos]
for class_videos in videos]
responses.append(sim_rsps)
return responses
def get_simulated_responses(model_dir, videos):
responses = []
for i in range(11):
with open(os.path.join(model_dir, 'mlp_%d' % i), 'rb') as f:
Wh, bh, Wo, bo = pickle.load(f)
mlp = create_mlp_from_params(Wh, bh, Wo, bo, rectify=True)
sim_rsps = [[transform(mlp, video) for video in class_videos]
for class_videos in videos]
responses.append(sim_rsps)
return responses
# Read videos.
# videos is a generator. So have to create it again and again.
print 'Reading videos...'
if stim_type == 'grating':
videos = read_grating_videos(grating_dirs)
else:
videos = read_videos(video_dir, spatial_downsample_factor)
print 'Applying the neural network (%d) transformation...' % i
"""
net.print_parameters()
net_features = [[transform(net, video, mean_rsps[i])
for video in class_videos]
for class_videos in videos]
"""
net_features = [[transform(net, video)
for video in class_videos]
for class_videos in videos]
print 'Dumping the neural features...'
net_features_dir = os.path.join(features_dir, 'net%d' % i)
dump_features(net_features, net_features_dir)
features.append(net_features)
predictions = []
for i, net_features in enumerate(features):
print 'Using features from network %d...' % i
if stim_type == 'grating':
tr_features, te_features = train_test_split_grating(net_features,
grating_train_idxs)