shanghaitech_dl.py

''' This config file will handle the video anomaly detection with dictionary learning (dl) '''

from ..base import *
from munch import DefaultMunch

dataset = 'shanghaitech'
modality = 'taskaware'
univ_data = 'kinetics400'

data_file_train = f'data/{dataset}/{dataset}.training.csv' # video list
data_file_test = f'data/{dataset}/{dataset}.testing.csv' # video list
ann_file_test = f'data/{dataset}/{dataset}_ground_truth.testing.json'

if 'universal' in modality:
    univ_dict_file = f'dictionary/{univ_data}/{univ_data}_dictionaries.{modality}.omp.100iters.npy'
    task_dict_file = None
    tmp_dict_file = f'dictionary/{dataset}/{dataset}_states.{modality}.npy'
elif 'taskaware' in modality:
    univ_dict_file = None
    task_dict_file = f'dictionary/{dataset}/{dataset}_dictionaries.{modality}.omp.100iters.90pct.npy'
    tmp_dict_file = None
elif 'univ' in modality and 'task' in modality:
    univ_dict_file = f'dictionary/{univ_data}/{univ_data}_dictionaries.universal.omp.100iters.npy'
    task_dict_file = f'dictionary/{dataset}/{dataset}_dictionaries.{modality}.omp.100iters.90pct.npy'
    tmp_dict_file = f'dictionary/{dataset}/{dataset}_states.universal.npy'

regular_file = f'dictionary/{dataset}/{dataset}_regular_features-2048dim.training.pickle'
random_state = 823
init_lr = 0.001

# dataset settings
base_dict = dict(
    dataset=dataset,
    data_root=data_root,
    backbone=backbone,
    quantize_size=quantize_size,
    dictionary=None,
    univ_dict_file=univ_dict_file,
    task_dict_file=task_dict_file,
    regular_file=regular_file,
    data_file=None,
    ann_file=None,
    tmp_dict_file=tmp_dict_file,
    modality=modality,
    dictionary_root=dictionary_root)

train_regular_dict = base_dict.copy()
train_anomaly_dict = base_dict.copy()
test_dict = base_dict.copy()

train_regular_dict.update(dict(test_mode=False, is_normal=True))
train_anomaly_dict.update(dict(test_mode=False, is_normal=False))
test_dict.update(dict(test_mode=True, is_normal=False))

data = dict(
    train=dict(
        regular=train_regular_dict,
        anomaly=train_anomaly_dict),
    test=test_dict)

data = DefaultMunch.fromDict(data)

data.train.regular.dataset = dataset
data.train.anomaly.dataset = dataset
data.test.dataset = dataset

data.train.regular.data_file = data_file_train
data.train.anomaly.data_file = data_file_train
data.test.data_file = data_file_test
data.test.ann_file = ann_file_test