def eval_metrics(cfg, metric):
with open(cfg) as fd:
data_specs = json.load(fd)
temp_size = data_specs['temp_size']
r = get_images_classes(
data_specs['images_path'],
data_specs['info_dict'],
data_specs['class_of_interest']
)
train_d, val_d, test_d = split_data(
data_specs['positive_ev_path'],
r,
window_size=temp_size,
future_size=0,
shuffle=False
)
train_d = add_texture_and_features_info(train_d, data_specs['info_path'])
val_d = add_texture_and_features_info(val_d, data_specs['info_path'])
test_d = add_texture_and_features_info(test_d, data_specs['info_path'])
full_df = pd.concat([train_d, val_d, test_d], ignore_index=True)
ds = ConcisePixelLevelDs(
full_df
)
if metric == 'AGES_frame':
ages_per_frame(ds)
if metric == 'pixels_AGE':
active_pixels_per_AGE(ds)
def get_single_plots(cfg):
with open(cfg) as fd:
data_specs = json.load(fd)
temp_size = data_specs['temp_size']
num_of_temp_features = data_specs['temp_features']
m = load_model(data_specs['model_path'],
# ToNameNet.name(),
DOTNetCNN.name(),
"model_params.json")
r = get_images_classes(
data_specs['images_path'],
data_specs['info_dict'],
data_specs['class_of_interest']
)
train_d, val_d, test_d = split_data(
data_specs['positive_ev_path'],
r,
window_size=temp_size,
future_size=0,
shuffle=True
)
info_path = data_specs["info_path"]
ds = PixelLevelDs(
val_d,
info_path=info_path,
add_polarity="neg" if num_of_temp_features > 1 else "",
)
plot_4D_predictions(m, ds, max_rows=5, black_out_aps=False)
def plot_transforms(cfg, apply_transform, max_rows=5):
with open(cfg) as fd:
data_specs = json.load(fd)
# temp_size = data_specs['temp_size']
temp_size = 3
num_of_temp_features = 1
individual_transforms = DelayedTransformSet(0.7,
transforms=[
DelayedRotation(0.6,
20,
expand=1),
CustomResize(500),
transforms.CenterCrop(
(260, 346))
])
r = get_images_classes(data_specs['images_path'], data_specs['info_dict'],
data_specs['class_of_interest'])
train_d, val_d, test_d = split_data(data_specs['positive_ev_path'],
r,
window_size=temp_size,
future_size=0,
shuffle=True)
info_path = data_specs["info_path"]
ds = PixelLevelDs(test_d,
info_path=info_path,
add_polarity="neg" if num_of_temp_features > 1 else "",
transforms=apply_transform,
individual_transforms=individual_transforms)
imgs_r, temp_data, ys = [], [], []
for j, (x, y) in enumerate(ds):
if j >= max_rows:
break
img_raw, temporal_data = x
imgs_r.append(img_raw[0].view(1, *img_raw[0].shape).numpy())
temp_data.append(temporal_data.numpy())
ys.append(y.numpy())
grid_plot(max_rows, 1, imgs_r, title="Static images")
grid_plot(max_rows,
temp_data[0].shape[0],
temp_data,
title="Net temporal input")
grid_plot(max_rows, ys[0].shape[0], ys, title="Real")
def load_data(cfg_list):
temp_size = 5
future_size = 0
val_size, test_size = .15, 0.30
train_list, val_list, test_list = [], [], []
for cfg in cfg_list:
with open(cfg) as fd:
data_specs = json.load(fd)
r = get_images_classes(
data_specs['images_path'],
data_specs['info_dict'],
data_specs['class_of_interest']
)
train_split, val_split, test_split = split_data(
data_specs['positive_ev_path'],
r,
window_size=temp_size,
future_size=future_size,
test_size=test_size,
val_size=val_size
)
train_split = add_texture_and_features_info(train_split, data_specs['info_path'])
val_split = add_texture_and_features_info(val_split, data_specs['info_path'])
test_split = add_texture_and_features_info(test_split, data_specs['info_path'])
train_list.append(train_split)
val_list.append(val_split)
test_list.append(test_split)
train_d, val_d, test_d = pd.concat(train_list), pd.concat(val_list), pd.concat(test_list)
train_d = train_d.sample(frac=1, replace=False, random_state=42).reset_index(drop=True)
val_d = val_d.sample(frac=1, replace=False, random_state=42).reset_index(drop=True)
test_d = test_d.sample(frac=1, replace=False, random_state=42).reset_index(drop=True)
return train_d, val_d, test_d
def make_sequence_video(cfg):
with open(cfg) as fd:
data_specs = json.load(fd)
temp_size = data_specs['temp_size']
num_of_temp_features = data_specs['temp_features']
m = load_model(data_specs['model_path'],
DOTNetCNN.name(),
# ToNameNet.name(),
"model_params.json")
use_gpu = torch_cuda.is_available()
device = torch_device(torch_cuda.current_device()) if use_gpu else torch_device("cpu")
m.to(device)
r = get_images_classes(
data_specs['images_path'],
data_specs['info_dict'],
data_specs['class_of_interest']
)
train_d, val_d, test_d = split_data(
data_specs['positive_ev_path'],
r,
window_size=temp_size,
future_size=0,
shuffle=False
)
full_df = pd.concat([train_d, val_d, test_d], ignore_index=True)
info_path = data_specs["info_path"]
ds = PixelLevelDs(
full_df,
info_path=info_path,
add_polarity="neg" if num_of_temp_features > 1 else "",
)
# make_video(m, ds)
make_non_repeating_video(m, ds)
def main():
temp_size = 5
future_size = 0
net_temp_output = future_size + 1
out_dims = (260, 346)
batch_size = 12
num_classes = 3
# use_gpu = torch_cuda.is_available()
use_gpu = False # pytorch is giving a memory error
device = torch_device(torch_cuda.current_device()) if use_gpu else torch_device("cpu")
with open("cfg.txt") as fd:
data_specs = json.load(fd)
r = get_images_classes(
data_specs['images_path'],
data_specs['info_dict'],
data_specs['class_of_interest']
)
train_d, val_d, test_d = split_data(
data_specs['positive_ev_path'],
r,
window_size=temp_size,
future_size=future_size
)
info_path = data_specs["info_path"]
train_ds = PixelLevelDs(
train_d,
info_path=info_path
)
val_ds = PixelLevelDs(
val_d,
# add_polarity="neg",
info_path=info_path
)
test_ds = PixelLevelDs(
test_d,
info_path=info_path
)
dls = {
"train": DataLoader(train_ds, batch_size=batch_size, num_workers=0, pin_memory=use_gpu),
"val": DataLoader(val_ds, batch_size=batch_size, num_workers=0, pin_memory=use_gpu),
"test": DataLoader(test_ds, batch_size=1)
}
full_forward = SaveStatsForwardCallback(heatmap_CE_accuracy(num_classes))
# to_device = ToDeviceCallback(device)
compose_callbacks = ComposeCallback(
start_callbacks=[StartPrinterCallback()],
train_pData_callbacks=[
# to_device,
full_forward],
val_pData_callbacks=[
# to_device,
full_forward],
nextEpoch_callbacks=[full_forward],
phaseEnded_callbacks=[full_forward],
allEnded_callbacks=[]
)
# TODO: read <A Closer Look at Spatiotemporal Convolutions for Action Recognition>
# https://arxiv.org/pdf/1711.11248.pdf
model = ToNameNet(temp_size, net_temp_output, out_dims, num_classes)
model.to(device)
train(model, dls['train'], dls['val'], compose_callbacks,
# criterion=heatmap_mse(num_classes),
criterion=heatmap_cross_entropy(num_classes, weight=tensor([0.3, 4, 15], dtype=torch_float)),
epochs=1)
def plot_on_concise_dataset(cfg_list, apply_transform, max_rows=5):
individual_transforms = DelayedTransformSet(0.7,
transforms=[
DelayedRotation(0.6,
20,
expand=1),
CustomResize(500),
transforms.CenterCrop(
(260, 346))
])
train_list, val_list, test_list = [], [], []
for cfg in cfg_list:
with open(cfg) as fd:
data_specs = json.load(fd)
r = get_images_classes(data_specs['images_path'],
data_specs['info_dict'],
data_specs['class_of_interest'])
train_split, val_split, test_split = split_data(
data_specs['positive_ev_path'],
r,
window_size=5,
future_size=1,
test_size=.3,
val_size=.2)
train_split = add_texture_and_features_info(train_split,
data_specs['info_path'])
val_split = add_texture_and_features_info(val_split,
data_specs['info_path'])
test_split = add_texture_and_features_info(test_split,
data_specs['info_path'])
train_list.append(train_split)
val_list.append(val_split)
test_list.append(test_split)
train_d, val_d, test_d = pd.concat(train_list), pd.concat(
val_list), pd.concat(test_list)
train_d = train_d.sample(frac=1, replace=False).reset_index(drop=True)
val_d = val_d.sample(frac=1, replace=False).reset_index(drop=True)
test_d = test_d.sample(frac=1, replace=False).reset_index(drop=True)
ds = ConcisePixelLevelDs(val_d,
transforms=apply_transform,
individual_transforms=individual_transforms)
imgs_r, temp_data, ys = [], [], []
for j, (x, y) in enumerate(ds):
if j >= max_rows:
break
img_raw, temporal_data = x
imgs_r.append(img_raw[0].view(1, *img_raw[0].shape).numpy())
temp_data.append(temporal_data.numpy())
ys.append(y.numpy())
grid_plot(max_rows, 1, imgs_r, title="Static images")
grid_plot(max_rows,
temp_data[0].shape[0],
temp_data,
title="Net temporal input")
grid_plot(max_rows, ys[0].shape[0], ys, title="Real")
def eval_metrics(cfg_list, metric, coi=2):
future_size = 0
val_size, test_size = .15, 0.30
# with open(cfg) as fd:
with open(cfg_list[0]) as fd:
data_specs = json.load(fd)
temp_size = data_specs['temp_size']
num_of_temp_features = data_specs['temp_features']
m = load_model(data_specs['model_path'],
DOTNetCNN.name(),
# ToNameNet.name(),
"model_params.json")
use_gpu = torch_cuda.is_available()
device = torch_device(torch_cuda.current_device()) if use_gpu else torch_device("cpu")
m.to(device)
# this for normal evaluations
# r = get_images_classes(
# data_specs['images_path'],
# data_specs['info_dict'],
# data_specs['class_of_interest']
# )
# train_d, val_d, test_d = split_data(
# data_specs['positive_ev_path'],
# r,
# window_size=temp_size,
# future_size=0,
# shuffle=False
# )
# train_d = add_texture_and_features_info(train_d, data_specs['info_path'])
# val_d = add_texture_and_features_info(val_d, data_specs['info_path'])
# test_d = add_texture_and_features_info(test_d, data_specs['info_path'])
# full_df = pd.concat([train_d, val_d, test_d], ignore_index=True)
# ds = ConcisePixelLevelDs(
# full_df
# )
train_list, val_list, test_list = [], [], []
for cfg in cfg_list:
with open(cfg) as fd:
data_specs = json.load(fd)
r = get_images_classes(
data_specs['images_path'],
data_specs['info_dict'],
data_specs['class_of_interest']
)
train_split, val_split, test_split = split_data(
data_specs['positive_ev_path'],
r,
window_size=temp_size,
future_size=future_size,
test_size=test_size,
val_size=val_size
)
train_split = add_texture_and_features_info(train_split, data_specs['info_path'])
val_split = add_texture_and_features_info(val_split, data_specs['info_path'])
test_split = add_texture_and_features_info(test_split, data_specs['info_path'])
train_list.append(train_split)
val_list.append(val_split)
test_list.append(test_split)
train_d, val_d, test_d = pd.concat(train_list), pd.concat(val_list), pd.concat(test_list)
train_d = train_d.sample(frac=1, replace=False, random_state=42).reset_index(drop=True)
val_d = val_d.sample(frac=1, replace=False, random_state=42).reset_index(drop=True)
test_d = test_d.sample(frac=1, replace=False, random_state=42).reset_index(drop=True)
ds = ConcisePixelLevelDs(
val_d
)
if metric == "event_acc":
return avg_accuracy_by_events(m, ds)
if metric == "ts":
return T_metrics_by_events(m, ds, class_of_interest=coi)
if metric == "yolo_IOU":
return get_bbox_IOU(m, ds, data_specs['yolo_path'])
if metric == "net_time":
return get_net_processing_time(m, ds)
if metric == "net_by_branch_time":
return get_processing_time_by_branch_separation(m, ds)
if metric == "pixel_T":
return T_metrics_by_pixels(m, ds, class_of_interest=coi, pool_k=10, pool_s=1)