class TensorboardLogs:
def __init__(self,
filepath: Union[Path, str],
num_scalars=0,
num_images=0):
size_guidance = {
SCALARS: num_scalars,
IMAGES: num_images,
}
self._event_acc = EventAccumulator(str(filepath),
size_guidance=size_guidance)
self._event_acc.Reload()
self.tags = self._event_acc.Tags()
self.all_tags = [
tag for _, some_tags in self.tags.items()
if isinstance(some_tags, list) for tag in some_tags
]
if len(self.all_tags) == 0:
warnings.warn(
'No tags detected, the path might not contain any logs')
def images(self, tag: str) -> List[Image.Image]:
return [
Image.open(io.BytesIO(img_event.encoded_image_string))
for img_event in self._event_acc.Images(tag)
]
def scalars(self, tag: str) -> torch.Tensor:
return torch.tensor([
scalar_event.value for scalar_event in self._event_acc.Scalars(tag)
])
def extract_data(summary_path: os.PathLike,
logs_path: os.PathLike) -> Tuple[List[float], Dict, str]:
config = Config.load(summary_path).config
events_acc = EventAccumulator(logs_path)
events_acc.Reload()
_, _, val_acc_diffs = zip(*events_acc.Scalars('diff/val/acc'))
hp = config.hp.to_dict()
hp['n_conv_layers'] = len(config.hp.conv_model_config.conv_sizes)
if 'Minimum_test' in events_acc.images.Keys():
image_test = events_acc.Images('Minimum_test')[0].encoded_image_string
image_train = events_acc.Images(
'Minimum_train')[0].encoded_image_string
else:
image_test = None
image_train = None
return val_acc_diffs, hp, image_test, image_train
def plot_alignment(log_dir, save_dir):
tf_event = glob.glob(os.path.join(log_dir, "events.*"))[0]
event_acc = EventAccumulator(tf_event, size_guidance=SIZE_GUIDANCE)
event_acc.Reload()
alignments = event_acc.Images('alignment')
for alignment in alignments:
img_str = alignment.encoded_image_string
step = alignment.step
f = io.BytesIO(img_str)
img = plt.imread(f)
plt.imshow(img)
plt.axis('off')
plt.title('step {:0>5d}'.format(step))
plt.tight_layout()
plt.savefig('{}/{:0>5d}.png'.format(save_dir, step))
plt.clf()
import os
import glob
import numpy as np
from scipy import misc
from tensorboard.backend.event_processing.event_accumulator import EventAccumulator
EVAL_DIR = './tmp/20170906-17:26:07/eval'
IMAGE_SHAPE = [256, 256]
event_acc = EventAccumulator(EVAL_DIR)
event_acc.Reload()
image_tags = event_acc.Tags()['images']
total = len(image_tags)
for i, tag in enumerate(image_tags):
encoded_image_string = event_acc.Images(tag)[0].encoded_image_string
ty = tag.split('/')[0]
index = tag.split('/')[2]
output_dir = os.path.join(EVAL_DIR, ty)
if not os.path.exists(output_dir):
os.makedirs(output_dir)
with open('%s/%s.png' % (output_dir, index), 'wb') as f:
f.write(encoded_image_string)
print '%d/%d exporting done.' % (i, total)
types = [
name for name in os.listdir(EVAL_DIR)
if os.path.isdir(os.path.join(EVAL_DIR, name))
]
n_row = 8
'/mnt/fs2/2019/Takamuro/m2_research/weather_transfer/results/c_UNet/tensorboard'
)
parser.add_argument(
'--log_path',
type=str,
default=
'runs/Apr11_20-10-29_DLBOX2_lr-0.0001_bs-16_ne-50_name-cUNet_w-e-res101-0408_train-D1T1_adam-b1-09_wloss-mse_train200k-test500'
)
args = parser.parse_args()
sys.path.append(os.getcwd())
output_path = os.path.join(args.output_dir, args.log_path.split('/')[-1])
os.makedirs(output_path, exist_ok=True)
path = args.log_path # Tensorboard ログファイル
event_acc = EventAccumulator(path, size_guidance={'images': 0})
event_acc.Reload() # ログファイルのサイズによっては非常に時間がかかる
for tag in event_acc.Tags()['images']:
events = event_acc.Images(tag)
tag_name = tag.replace('/', '_')
for index, event in enumerate(events):
# 画像はエンコードされているので戻す
s = np.frombuffer(event.encoded_image_string, dtype=np.uint8)
image = cv2.imdecode(s, cv2.IMREAD_COLOR) # カラー画像の場合
# 保存
output_path_ = os.path.join(output_path,
'{}_{:04}.jpg'.format(tag_name, index))
cv2.imwrite(output_path_, image)
