feat(detection): support time and isort (#42)

.isort.cfg

+[isort]
+line_length = 100
+multi_line_output = 3
+balanced_wrapping = True
+# skip = tools/test_net.py, tools/train_net.py
+known_standard_library = setuptools
+known_myself = official
+known_data_processing = cv2,numpy,scipy,PIL,matplotlib
+known_datasets = pycocotools
+known_deeplearning = megengine
+sections = FUTURE,STDLIB,THIRDPARTY,FIRSTPARTY,data_processing,datasets,deeplearning,myself,LOCALFOLDER
+no_lines_before=STDLIB,THIRDPARTY,datasets
+default_section = FIRSTPARTY

official/vision/detection/layers/basic/functional.py

@@ -6,10 +6,10 @@
 # Unless required by applicable law or agreed to in writing,
 # software distributed under the License is distributed on an
 # "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-import megengine as mge
-import megengine.functional as F
 import numpy as np
 
+import megengine as mge
+import megengine.functional as F
 from megengine.core import Tensor
 
 

official/vision/detection/layers/basic/norm.py

@@ -24,9 +24,9 @@
 # This file has been modified by Megvii ("Megvii Modifications").
 # All Megvii Modifications are Copyright (C) 2014-2020 Megvii Inc. All rights reserved.
 # ---------------------------------------------------------------------
-import megengine.module as M
 import numpy as np
 
+import megengine.module as M
 from megengine.core import Buffer
 
 

official/vision/detection/layers/det/anchor.py

@@ -8,10 +8,10 @@
 # "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 from abc import ABCMeta, abstractmethod
 
-import megengine.functional as F
 import numpy as np
 
-from megengine.core import tensor, Tensor
+import megengine.functional as F
+from megengine.core import Tensor, tensor
 
 
 class BaseAnchorGenerator(metaclass=ABCMeta):

official/vision/detection/layers/det/box_utils.py

@@ -9,7 +9,6 @@
 from abc import ABCMeta, abstractmethod
 
 import megengine.functional as F
-
 from megengine.core import Tensor
 
 

official/vision/detection/layers/det/loss.py

@@ -7,7 +7,6 @@
 # software distributed under the License is distributed on an
 # "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 import megengine.functional as F
-
 from megengine.core import Tensor
 
 

official/vision/detection/layers/det/pooler.py

@@ -9,6 +9,7 @@
 import math
 
 import numpy as np
+
 import megengine as mge
 import megengine.functional as F
 

official/vision/detection/layers/det/retinanet.py

@@ -10,7 +10,6 @@ import math
 from typing import List
 
 import megengine.module as M
-
 from megengine.core import Tensor
 
 from official.vision.detection.layers import basic

official/vision/detection/layers/det/rpn.py

@@ -7,9 +7,10 @@
 # software distributed under the License is distributed on an
 # "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 import megengine as mge
-import megengine.random as rand
 import megengine.functional as F
 import megengine.module as M
+import megengine.random as rand
+
 from official.vision.detection import layers
 from official.vision.detection.tools.gpu_nms import batched_nms
 

official/vision/detection/models/faster_rcnn_fpn.py

@@ -7,6 +7,7 @@
 # software distributed under the License is distributed on an
 # "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 import numpy as np
+
 import megengine as mge
 import megengine.functional as F
 import megengine.module as M

official/vision/detection/models/retinanet.py

@@ -6,10 +6,11 @@
 # Unless required by applicable law or agreed to in writing,
 # software distributed under the License is distributed on an
 # "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+import numpy as np
+
 import megengine as mge
 import megengine.functional as F
 import megengine.module as M
-import numpy as np
 
 from official.vision.classification.resnet.model import resnet50
 from official.vision.detection import layers

official/vision/detection/tools/gpu_nms.py

@@ -5,6 +5,7 @@ import os
 import struct
 
 import numpy as np
+
 import megengine as mge
 import megengine.functional as F
 from megengine._internal.craniotome import CraniotomeBase

official/vision/detection/tools/inference.py

@@ -12,8 +12,9 @@ import os
 import sys
 
 import cv2
-import megengine as mge
 import numpy as np
+
+import megengine as mge
 from megengine import jit
 from megengine.data.dataset import COCO
 

official/vision/detection/tools/test.py

@@ -13,13 +13,14 @@ import os
 import random
 import sys
 from multiprocessing import Process, Queue
+from tqdm import tqdm
 
 import cv2
-import megengine as mge
 import numpy as np
+
+import megengine as mge
 from megengine import jit
 from megengine.data import DataLoader, SequentialSampler
-from tqdm import tqdm
 
 from official.vision.detection.tools.data_mapper import data_mapper
 from official.vision.detection.tools.nms import py_cpu_nms

official/vision/detection/tools/train.py

@@ -14,16 +14,18 @@ import importlib
 import multiprocessing as mp
 import os
 import sys
+import time
 from collections import defaultdict
+from tabulate import tabulate
 
-import megengine as mge
 import numpy as np
+
+import megengine as mge
 from megengine import distributed as dist
 from megengine import jit
 from megengine import optimizer as optim
 from megengine.data import Collator, DataLoader, Infinite, RandomSampler
 from megengine.data import transform as T
-from tabulate import tabulate
 
 from official.vision.detection.tools.data_mapper import data_mapper
 
@@ -81,6 +83,7 @@ def worker(rank, world_size, args):
         weights = mge.load(args.weight_file)
         model.backbone.bottom_up.load_state_dict(weights)
 
+    if rank == 0:
         logger.info("Prepare dataset")
     loader = build_dataloader(model.batch_size, args.dataset_dir, model.cfg)
     train_loader = iter(loader["train"])
@@ -155,21 +158,32 @@ def train_one_epoch(
         return losses
 
     meter = AverageMeter(record_len=model.cfg.num_losses)
+    time_meter = AverageMeter(record_len=2)
     log_interval = model.cfg.log_interval
     for step in range(tot_steps):
         adjust_learning_rate(opt, epoch_id, step, model, world_size)
+
+        data_tik = time.time()
         mini_batch = next(data_queue)
+        data_tok = time.time()
+
         model.inputs["image"].set_value(mini_batch["data"])
         model.inputs["gt_boxes"].set_value(mini_batch["gt_boxes"])
         model.inputs["im_info"].set_value(mini_batch["im_info"])
 
+        tik = time.time()
         opt.zero_grad()
         loss_list = propagate()
         opt.step()
+        tok = time.time()
+
+        time_meter.update([tok - tik, data_tok - data_tik])
 
         if rank == 0:
+            info_str = "e%d, %d/%d, lr:%f, "
             loss_str = ", ".join(["{}:%f".format(loss) for loss in model.cfg.losses_keys])
-            log_info_str = "e%d, %d/%d, lr:%f, " + loss_str
+            time_str = ", train_time:%.3fs, data_time:%.3fs"
+            log_info_str = info_str + loss_str + time_str
             meter.update([loss.numpy() for loss in loss_list])
             if step % log_interval == 0:
                 average_loss = meter.average()
@@ -180,8 +194,10 @@ def train_one_epoch(
                     tot_steps,
                     opt.param_groups[0]["lr"],
                     *average_loss,
+                    *time_meter.average()
                 )
                 meter.reset()
+                time_meter.reset()
 
 
 def make_parser():