def get_data(config, split):
data = CachedNTUGems()
indices = get_split_indices(data)[split]
data = SubDataset(data, indices)
# apply generic preprocessing (in particular resizing and masking)
data = ProcessedDataset(
data, TargetProcessing(config.get("spatial_size", 256), 1))
# crop images
data = ProcessedDataset(data, center_crop)
# add required labels
data = ExtraLabelsDataset(data, add_labels)
return data
def __init__(self, root, show_bar=False):
er = EvalReader(root)
if "model_output.csv" not in root:
csv_path = os.path.join(root, "model_output.csv")
else:
csv_path = root
root = os.path.dirname(root)
self.labels = load_labels(os.path.join(root, "model_outputs"))
# Capture the case that only labels have been written out
try:
csv_data = CsvDataset(csv_path, comment="#")
self.data = ProcessedDataset(csv_data, er)
except pd.errors.EmptyDataError as e:
exemplar_labels = self.labels[sorted(self.labels.keys())[0]]
self.data = EmptyDataset(len(exemplar_labels), self.labels)
self.append_labels = True
if 'model_output.csv' in root:
root = root[:-len('model_output.csv')]
savename_score = os.path.join(save_dir, 'score.txt')
#savename_std = os.path.join(save_dir, 'std.txt')
fid_score = np.array(fids).mean()
#fid_std = np.array(fids).std()
with open(savename_score, 'w+') as f:
f.write(str(fid_score))
#with open(savename_std, 'w+') as f:
# f.write(str(fid_std))
print('\nFID SCORE: {:.2f}'.format(fid_score))
return {"scalars": {"fid": fid_score}}
if __name__ == "__main__":
from edflow.debug import DebugDataset
from edflow.data.dataset import ProcessedDataset
D1 = DebugDataset(size=100)
D2 = DebugDataset(size=100)
P = lambda *args, **kwargs: {'image': np.ones([256, 256, 3])}
D1 = ProcessedDataset(D1, P)
D2 = ProcessedDataset(D2, P)
print(D1[0])
fid('.', D1, D2, {})