def _start_dataset_transform(self,
dataset_dicts,
dataset_tranforms,
pre_context_dict=None):
""" 总体逻辑，对于所有的transform依次执行：
pre_process 所有的数据
process 单条数据，并且进行filter
然后将得到的dds输入到下一个管道中
if runtime == True : means __getitem__() function, can't return None
if runtime == False: means _load_annotations() function
"""
timer = Timer()
context = DatasetTransformContext()
if pre_context_dict:
for k, v in pre_context_dict.items():
context[k] = v
for transform in dataset_tranforms:
#print (dataset_dicts[0])
transform.pre_process(dataset_dicts, context)
new_dataset_dicts = []
for dd in dataset_dicts:
if dd and transform._runtime == False:
dd = transform.process(dd, context)
if dd: new_dataset_dicts.append(dd)
dataset_dicts = new_dataset_dicts
logging.info("DatasetTransform {} takes {:.2f} seconds.".format(
self.raw_name, timer.seconds()))
return dataset_dicts, context
def _load_annotations(self):
timer = Timer()
"""Constructs the imdb."""
# Compile the split data path
logger.info('{} data path: {}'.format(self.name, self.label_file))
# Construct the image db
imdb = []
f = open(self.label_file, "r")
for line in f.readlines():
img_path, label = line.strip().split(" ")
imdb.append({
"im_path": os.path.join(self.image_root, img_path),
"class": int(label),
})
f.close()
logging.info("Loading {} takes {:.2f} seconds.".format(
self.label_file, timer.seconds()))
dataset_dicts = []
for i, item in enumerate(imdb):
dataset_dicts.append({
"image_id": i,
"category_id": item["class"],
"file_name": item["im_path"],
})
return dataset_dicts
def _load_clean(self):
timer = Timer()
def sample_clean(matrix, clean_num):
mat = matrix.reshape([-1])
clean_mat = np.zeros_like(mat)
import random
random.seed(self.seed)
pool = set()
while (len(pool) != clean_num):
ind = random.choice(range(mat.size))
if ind not in pool:
clean_mat[ind] = mat[ind]
pool.add(ind)
print(list(pool)[:10])
assert (clean_mat != 0).sum() == clean_num, "Error"
return clean_mat
if self.dataset_type == "train":
path = osp.join(self.dataset_root, "select_bias",
"matrix_after.npy")
matrix = np.load(path)
matrix = sample_clean(matrix, self.clean_num)
matrix = matrix.reshape([self.user_num, self.item_num])
dataset_dicts = self._matrix2datasets(matrix)
else:
path = osp.join(self.dataset_root, "select_bias",
"clean_matrix.npy")
dataset_dicts = self._load_matrix(path)
logging.info("Loading MovieLen ::{} takes {:.2f} seconds.".format(
"movie_len_noise", timer.seconds()))
return dataset_dicts
def benchmark_eval(args):
cfg = setup(args)
model = build_model(cfg)
model.eval()
logger.info("Model:\n{}".format(model))
DetectionCheckpointer(model).load(cfg.MODEL.WEIGHTS)
cfg.defrost()
cfg.DATALOADER.NUM_WORKERS = 0
data_loader = build_detection_test_loader(cfg, cfg.DATASETS.TEST[0])
dummy_data = list(itertools.islice(data_loader, 100))
def f():
while True:
yield from DatasetFromList(dummy_data, copy=False)
for _ in range(5): # warmup
model(dummy_data[0])
max_iter = 400
timer = Timer()
with tqdm.tqdm(total=max_iter) as pbar:
for idx, d in enumerate(f()):
if idx == max_iter:
break
model(d)
pbar.update()
logger.info("{} iters in {} seconds.".format(max_iter, timer.seconds()))
def _load_true(self):
timer = Timer()
path = osp.join(self.dataset_root, "select_bias",
"matrix_after" + ".npy")
dataset_dicts = self._load_matrix(path)
logging.info("Loading MovieLen ::{} takes {:.2f} seconds.".format(
"movie_len_true", timer.seconds()))
return dataset_dicts
def _load_annotations(self):
timer = Timer()
dataloader = self.dataloader = DataLoader(
osp.join(self.dataset_root, self.dataset_name), self.with_feature)
dataloader.generate_pair_wise_training_corpus()
logging.info("Loading AmazonDataset::{} takes {:.2f} seconds.".format(
self.dataset_name, timer.seconds()))
dataset_dicts = []
dataset_statistic = dataloader.statistics
if (self.is_train): # fill the dataset_dicts
if (not self.with_feature):
for user, pos_item, neg_item in zip(
dataloader.user_all,
dataloader.pos_item_all,
dataloader.neg_item_all,
):
dataset_dicts.append({
"user": user,
#"feat": feat,
"pos_item": pos_item,
#"pos_feat": pos_feat,
"neg_item": neg_item,
#"neg_feat": neg_feat,
})
elif (self.with_feature):
for user, pos_item, neg_item, feat, x_uf, x_if in zip(
dataloader.user_all, dataloader.pos_item_all,
dataloader.neg_item_all, dataloader.pos_feature_all,
dataloader.x_uf, dataloader.x_if):
dataset_dicts.append({
"user": user,
"feat": feat,
"pos_item": pos_item,
"neg_item": neg_item,
"x_uf": x_uf,
"x_if": x_if
})
else: # fill the dataset_dicts with different logic
#assert len(dataloader.ground_truth_user_items_dict) == len(dataloader.compute_user_items_dict), "AmazonDataset have some wrong with it, len(gt)!=len(candidate)"
validate_user = set(
dataloader.ground_truth_user_items_dict.keys()) & set(
dataloader.compute_user_items_dict.keys())
for user in validate_user:
gt = dataloader.ground_truth_user_items_dict[user]
candidate = dataloader.compute_user_items_dict[user]
dataset_dicts.append({
"user": user,
"gt": gt,
"candidate": candidate,
})
return dataset_dicts, dataset_statistic
def _load_noise(self):
timer = Timer()
if self.dataset_type == "train":
path = osp.join(self.dataset_root, "select_bias",
"observe_matrix_" + str(self.alpha) + ".npy")
else:
path = osp.join(self.dataset_root, "select_bias",
"clean_matrix.npy")
dataset_dicts = self._load_matrix(path)
logging.info("Loading MovieLen ::{} takes {:.2f} seconds.".format(
"movie_len_noise", timer.seconds()))
return dataset_dicts
def benchmark_data(args):
cfg = setup(args)
dataloader = build_detection_train_loader(cfg)
timer = Timer()
itr = iter(dataloader)
for i in range(10): # warmup
next(itr)
if i == 0:
startup_time = timer.seconds()
timer = Timer()
max_iter = 1000
for _ in tqdm.trange(max_iter):
next(itr)
logger.info("{} iters ({} images) in {} seconds.".format(
max_iter, max_iter * cfg.SOLVER.IMS_PER_BATCH, timer.seconds()))
logger.info("Startup time: {} seconds".format(startup_time))
vram = psutil.virtual_memory()
logger.info("RAM Usage: {:.2f}/{:.2f} GB".format(
(vram.total - vram.available) / 1024**3, vram.total / 1024**3))
def _runtime_dataset_transform(self, dd, dataset_tranforms):
""" runtime stage transform proceduce
"""
timer = Timer()
context = self.context # reuse the global information
assert dd, "input dd is None, Some wrong happen in your dataset or you put None object in self.dataset_dict"
for transform in dataset_tranforms:
if transform._runtime == True:
dd = transform.process(dd, context)
assert dd, "return value can't be None, when in runtime mode"
logging.info(
"Runtime DatasetTrasnform {} takes {:.2f} seconds.".format(
self.raw_name, timer.seconds()))
self.context = context # reset
return dd
def _load_annotations(self):
timer = Timer()
self.noise_train, self.clean, self.noise_val, self.test, num_classes = generate_noisy_cifar(
self.dataset_name, self.dataset_root, self.validate_number,
self.noise_ratio, self.num_clean, self.seed, self.with_background)
dataset_statistic = {
'num_classes': num_classes,
'format': 'HWC',
}
self.type2data = {
'noise-train': self.noise_train,
'clean': self.clean,
'noise-val': self.noise_val,
'test': self.test,
}
dataset_dicts = []
current_img, current_label, current_mask = self.type2data[
self.data_type]
if self.dataset_name == 'cifar-10':
dataset_statistic['mean'] = (0.4914, 0.4822, 0.4465)
dataset_statistic['std'] = (0.2023, 0.1994, 0.2010)
elif self.dataset_name == 'cifar-100':
dataset_statistic['mean'] = (0.5071, 0.4867, 0.4408)
dataset_statistic['std'] = (0.2675, 0.2565, 0.2761)
else:
raise RuntimeError("name must be cifar-10 / cifar-100")
assert (len(current_img) == len(current_label))
assert (current_mask is None or len(current_img) == len(current_mask))
#print (current_img[0])
for i in range(len(current_img)):
item_dict = {}
item_dict['image'] = current_img[i] # H x W x C
item_dict['image_id'] = i # H x W x C
item_dict['category_id'] = current_label[i] # int
if current_mask is not None:
item_dict['is_clean'] = current_mask[i]
else:
item_dict['is_clean'] = 1
item_dict['width'] = 32
item_dict['height'] = 32
item_dict['channel'] = 3
dataset_dicts.append(item_dict)
tot_len = len(dataset_dicts)
dataset_dicts = dataset_dicts[:int(tot_len * self.discount)]
print("dataset length: ", len(dataset_dicts))
logging.info("Loading CIFAR Images::{} takes {:.2f} seconds.".format(
self.dataset_name, timer.seconds()))
return dataset_dicts, dataset_statistic
def _load_annotations(self):
timer = Timer()
raw_trainset, raw_testset = get_torch_mnist_dataset()
# every set is a tuple of (datas, labels)
__import__('pdb').set_trace()
train_set, val_set, test_set, train_pos_set, train_neg_set = get_imbalance_dataset(
raw_trainset, raw_testset)
dataset_statistic = {
'num_classes': 10,
'format': 'HWC',
}
self.type2data = {
'train': train_set,
'val': val_set,
'test': test_set,
}
__import__('pdb').set_trace()
dataset_dicts = []
current_img, current_label, current_mask = self.type2data[
self.data_type]
assert (len(current_img) == len(current_label))
assert (current_mask is None or len(current_img) == len(current_mask))
for i in range(len(current_img)):
item_dict = {}
item_dict['image'] = current_img[i] # H x W x C
item_dict['image_id'] = i # H x W x C
item_dict['category_id'] = current_label[i] # int
if current_mask is not None:
item_dict['is_clean'] = current_mask[i]
else:
item_dict['is_clean'] = 1
item_dict['width'] = 28
item_dict['height'] = 28
item_dict['channel'] = 1
dataset_dicts.append(item_dict)
tot_len = len(dataset_dicts)
dataset_dicts = dataset_dicts[:int(tot_len * self.discount)]
print("dataset length: ", len(dataset_dicts))
logging.info("Loading CIFAR Images::{} takes {:.2f} seconds.".format(
self.dataset_name, timer.seconds()))
return dataset_dicts, dataset_statistic
def _load_dataset(self):
timer = Timer()
if self.with_feature:
self._load_feature()
meta_path = osp.join(self.dataset_root, self.dataset_name, 'meta.txt')
with open(meta_path, "r") as fp:
lines = fp.readlines()
self.user_num = int(lines[0].strip())
self.item_num = int(lines[1].strip())
data_path = osp.join(self.dataset_root, self.dataset_name,
self.dataset_type + ".ascii")
prop_path = osp.join(self.dataset_root, self.dataset_name,
"propensities" + ".ascii")
self.mat_prop = None
if (osp.exists(prop_path)): self.mat_prop = np.loadtxt(prop_path)
dataset_dicts = self._load_matrix(data_path)
logging.info("Loading dataset ::{} takes {:.2f} seconds.".format(
self.dataset_name, timer.seconds()))
return dataset_dicts
def _load_annotations(self):
timer = Timer()
"""Constructs the imdb."""
# Compile the split data path
logger.info('{} data path: {}'.format(self.name, self.label_file))
# Images are stored per class in subdirs (format: n<number>)
class_ids = [k for k, v in IMAGENET_CATEGORIES.items()]
class_id_cont_id = {
k: v[0] - 1
for k, v in IMAGENET_CATEGORIES.items()
}
# class_ids = sorted([
# f for f in os.listdir(split_path) if re.match(r'^n[0-9]+$', f)
# ])
# # Map ImageNet class ids to contiguous ids
# class_id_cont_id = {v: i for i, v in enumerate(class_ids)}
# Construct the image db
imdb = []
if "://" not in self.image_root:
for class_id in class_ids:
cont_id = class_id_cont_id[class_id]
im_dir = os.path.join(self.label_file, class_id)
for im_name in os.listdir(im_dir):
imdb.append({
'im_path': os.path.join(im_dir, im_name),
'class': cont_id,
})
logging.info("Loading {} takes {:.2f} seconds.".format(
self.label_file, timer.seconds()))
dataset_dicts = []
for i, item in enumerate(imdb):
dataset_dicts.append({
"image_id": i,
"category_id": item["class"],
"file_name": item["im_path"],
})
return dataset_dicts
def _load_raw(self):
timer = Timer()
mmap = {
"train": "base",
"test": "test",
}
path = osp.join(
self.dataset_root,
"u" + str(self.num_fold) + "." + mmap[self.dataset_type])
with open(path, "r") as fp:
lines = fp.readlines()
dataset_dicts = []
for line in lines:
fields = line.strip().split("\t")
user = int(fields[0])
item = int(fields[1])
score = int(fields[2])
dataset_dicts.append((user, item, score))
logging.info("Loading MovieLen ::{} takes {:.2f} seconds.".format(
self.dataset_name, timer.seconds()))
return dataset_dicts
def _load_annotations(self, json_file, image_root):
"""
Load a json file in LVIS's annotation format.
Args:
json_file (str): full path to the LVIS json annotation file.
image_root (str): the directory where the images in this json file exists.
Returns:
list[dict]: a list of dicts in cvpods standard format. (See
`Using Custom Datasets </tutorials/datasets.html>`_ )
Notes:
1. This function does not read the image files.
The results do not have the "image" field.
"""
from lvis import LVIS
json_file = PathManager.get_local_path(json_file)
timer = Timer()
lvis_api = LVIS(json_file)
if timer.seconds() > 1:
logger.info("Loading {} takes {:.2f} seconds.".format(
json_file, timer.seconds()))
# sort indices for reproducible results
img_ids = sorted(lvis_api.imgs.keys())
# imgs is a list of dicts, each looks something like:
# {'license': 4,
# 'url': 'http://farm6.staticflickr.com/5454/9413846304_881d5e5c3b_z.jpg',
# 'file_name': 'COCO_val2014_000000001268.jpg',
# 'height': 427,
# 'width': 640,
# 'date_captured': '2013-11-17 05:57:24',
# 'id': 1268}
imgs = lvis_api.load_imgs(img_ids)
# anns is a list[list[dict]], where each dict is an annotation
# record for an object. The inner list enumerates the objects in an image
# and the outer list enumerates over images. Example of anns[0]:
# [{'segmentation': [[192.81,
# 247.09,
# ...
# 219.03,
# 249.06]],
# 'area': 1035.749,
# 'image_id': 1268,
# 'bbox': [192.81, 224.8, 74.73, 33.43],
# 'category_id': 16,
# 'id': 42986},
# ...]
anns = [lvis_api.img_ann_map[img_id] for img_id in img_ids]
# Sanity check that each annotation has a unique id
ann_ids = [
ann["id"] for anns_per_image in anns for ann in anns_per_image
]
assert len(set(ann_ids)) == len(
ann_ids), "Annotation ids in '{}' are not unique".format(json_file)
imgs_anns = list(zip(imgs, anns))
logger.info("Loaded {} images in the LVIS format from {}".format(
len(imgs_anns), json_file))
dataset_dicts = []
for (img_dict, anno_dict_list) in imgs_anns:
record = {}
file_name = img_dict["file_name"]
if img_dict["file_name"].startswith("COCO"):
# Convert form the COCO 2014 file naming convention of
# COCO_[train/val/test]2014_000000000000.jpg to the 2017 naming convention of
# 000000000000.jpg (LVIS v1 will fix this naming issue)
file_name = file_name[-16:]
record["file_name"] = os.path.join(image_root, file_name)
record["height"] = img_dict["height"]
record["width"] = img_dict["width"]
record["not_exhaustive_category_ids"] = img_dict.get(
"not_exhaustive_category_ids", [])
record["neg_category_ids"] = img_dict.get("neg_category_ids", [])
image_id = record["image_id"] = img_dict["id"]
objs = []
for anno in anno_dict_list:
# Check that the image_id in this annotation is the same as
# the image_id we're looking at.
# This fails only when the data parsing logic or the annotation file is buggy.
assert anno["image_id"] == image_id
obj = {"bbox": anno["bbox"], "bbox_mode": BoxMode.XYWH_ABS}
obj["category_id"] = anno[
"category_id"] - 1 # Convert 1-indexed to 0-indexed
segm = anno["segmentation"] # list[list[float]]
# filter out invalid polygons (< 3 points)
valid_segm = [
poly for poly in segm
if len(poly) % 2 == 0 and len(poly) >= 6
]
assert len(segm) == len(
valid_segm
), "Annotation contains an invalid polygon with < 3 points"
assert len(segm) > 0
obj["segmentation"] = segm
objs.append(obj)
record["annotations"] = objs
dataset_dicts.append(record)
return dataset_dicts
def _load_annotations(self, json_file, image_root):
"""
Load a json file with CrowdHuman's instances annotation format.
Currently supports instance detection, instance segmentation,
and person keypoints annotations.
Args:
json_file (str): full path to the json file in CrowdHuman instances annotation format.
image_root (str): the directory where the images in this json file exists.
Returns:
list[dict]: a list of dicts in cvpods standard format. (See
`Using Custom Datasets </tutorials/datasets.html>`_ )
Notes:
1. This function does not read the image files.
The results do not have the "image" field.
"""
timer = Timer()
json_file = PathManager.get_local_path(json_file)
with open(json_file, 'r') as file:
gt_records = file.readlines()
if timer.seconds() > 1:
logger.info("Loading {} takes {:.2f} seconds.".format(
json_file, timer.seconds()))
logger.info("Loaded {} images in CrowdHuman format from {}".format(
len(gt_records), json_file))
dataset_dicts = []
ann_keys = ["tag", "hbox", "vbox", "head_attr", "extra"]
for anno_str in gt_records:
anno_dict = json.loads(anno_str)
record = {}
record["file_name"] = os.path.join(
image_root, "{}.jpg".format(anno_dict["ID"]))
record["image_id"] = anno_dict["ID"]
objs = []
for anno in anno_dict['gtboxes']:
# Check that the image_id in this annotation is the same as
# the image_id we're looking at.
# This fails only when the data parsing logic or the annotation file is buggy.
# The original COCO valminusminival2014 & minival2014 annotation files
# actually contains bugs that, together with certain ways of using COCO API,
# can trigger this assertion.
obj = {key: anno[key] for key in ann_keys if key in anno}
obj["bbox"] = anno["fbox"]
obj["category_id"] = 0
if 'extra' in anno and 'ignore' in anno[
'extra'] and anno['extra']['ignore'] != 0:
obj["category_id"] = -1
obj["bbox_mode"] = BoxMode.XYWH_ABS
objs.append(obj)
record["annotations"] = objs
dataset_dicts.append(record)
return dataset_dicts
def _load_annotations(self,
json_file,
image_root,
dataset_name=None,
extra_annotation_keys=None):
"""
Load a json file with COCO's instances annotation format.
Currently supports instance detection, instance segmentation,
and person keypoints annotations.
Args:
json_file (str): full path to the json file in COCO instances annotation format.
image_root (str): the directory where the images in this json file exists.
dataset_name (str): the name of the dataset (e.g., coco_2017_train).
If provided, this function will also put "thing_classes" into
the metadata associated with this dataset.
extra_annotation_keys (list[str]): list of per-annotation keys that should also be
loaded into the dataset dict (besides "iscrowd", "bbox", "keypoints",
"category_id", "segmentation"). The values for these keys will be returned as-is.
For example, the densepose annotations are loaded in this way.
Returns:
list[dict]: a list of dicts in cvpods standard format. (See
`Using Custom Datasets </tutorials/datasets.html>`_ )
Notes:
1. This function does not read the image files.
The results do not have the "image" field.
"""
from pycocotools.coco import COCO
timer = Timer()
json_file = PathManager.get_local_path(json_file)
with contextlib.redirect_stdout(io.StringIO()):
coco_api = COCO(json_file)
if timer.seconds() > 1:
logger.info("Loading {} takes {:.2f} seconds.".format(
json_file, timer.seconds()))
id_map = None
if dataset_name is not None:
cat_ids = sorted(coco_api.getCatIds())
cats = coco_api.loadCats(cat_ids)
# The categories in a custom json file may not be sorted.
thing_classes = [
c["name"] for c in sorted(cats, key=lambda x: x["id"])
]
self.meta["thing_classes"] = thing_classes
# In COCO, certain category ids are artificially removed,
# and by convention they are always ignored.
# We deal with COCO's id issue and translate
# the category ids to contiguous ids in [0, 80).
# It works by looking at the "categories" field in the json, therefore
# if users' own json also have incontiguous ids, we'll
# apply this mapping as well but print a warning.
if not (min(cat_ids) == 1 and max(cat_ids) == len(cat_ids)):
if "coco" not in dataset_name:
logger.warning("""
Category ids in annotations are not in [1, #categories]! We'll apply a mapping for you.
""")
id_map = {v: i for i, v in enumerate(cat_ids)}
self.meta["thing_dataset_id_to_contiguous_id"] = id_map
# sort indices for reproducible results
img_ids = sorted(coco_api.imgs.keys())
# imgs is a list of dicts, each looks something like:
# {'license': 4,
# 'url': 'http://farm6.staticflickr.com/5454/9413846304_881d5e5c3b_z.jpg',
# 'file_name': 'COCO_val2014_000000001268.jpg',
# 'height': 427,
# 'width': 640,
# 'date_captured': '2013-11-17 05:57:24',
# 'id': 1268}
imgs = coco_api.loadImgs(img_ids)
# anns is a list[list[dict]], where each dict is an annotation
# record for an object. The inner list enumerates the objects in an image
# and the outer list enumerates over images. Example of anns[0]:
# [{'segmentation': [[192.81,
# 247.09,
# ...
# 219.03,
# 249.06]],
# 'area': 1035.749,
# 'iscrowd': 0,
# 'image_id': 1268,
# 'bbox': [192.81, 224.8, 74.73, 33.43],
# 'category_id': 16,
# 'id': 42986},
# ...]
anns = [coco_api.imgToAnns[img_id] for img_id in img_ids]
if "minival" not in json_file:
# The popular valminusminival & minival annotations for COCO2014 contain this bug.
# However the ratio of buggy annotations there is tiny and does not affect accuracy.
# Therefore we explicitly white-list them.
ann_ids = [
ann["id"] for anns_per_image in anns for ann in anns_per_image
]
assert len(set(ann_ids)) == len(
ann_ids), "Annotation ids in '{}' are not unique!".format(
json_file)
imgs_anns = list(zip(imgs, anns))
logger.info("Loaded {} images in COCO format from {}".format(
len(imgs_anns), json_file))
dataset_dicts = []
ann_keys = ["iscrowd", "bbox", "keypoints", "category_id"
] + (extra_annotation_keys or [])
num_instances_without_valid_segmentation = 0
for (img_dict, anno_dict_list) in imgs_anns:
record = {}
record["file_name"] = os.path.join(image_root,
img_dict["file_name"])
record["height"] = img_dict["height"]
record["width"] = img_dict["width"]
image_id = record["image_id"] = img_dict["id"]
objs = []
for anno in anno_dict_list:
# Check that the image_id in this annotation is the same as
# the image_id we're looking at.
# This fails only when the data parsing logic or the annotation file is buggy.
# The original COCO valminusminival2014 & minival2014 annotation files
# actually contains bugs that, together with certain ways of using COCO API,
# can trigger this assertion.
assert anno["image_id"] == image_id
assert anno.get("ignore", 0) == 0
obj = {key: anno[key] for key in ann_keys if key in anno}
segm = anno.get("segmentation", None)
if segm: # either list[list[float]] or dict(RLE)
if not isinstance(segm, dict):
# filter out invalid polygons (< 3 points)
segm = [
poly for poly in segm
if len(poly) % 2 == 0 and len(poly) >= 6
]
if len(segm) == 0:
num_instances_without_valid_segmentation += 1
continue # ignore this instance
obj["segmentation"] = segm
keypts = anno.get("keypoints", None)
if keypts: # list[int]
for idx, v in enumerate(keypts):
if idx % 3 != 2:
# COCO's segmentation coordinates are floating points in [0, H or W],
# but keypoint coordinates are integers in [0, H-1 or W-1]
# Therefore we assume the coordinates are "pixel indices" and
# add 0.5 to convert to floating point coordinates.
keypts[idx] = v + 0.5
obj["keypoints"] = keypts
obj["bbox_mode"] = BoxMode.XYWH_ABS
if id_map:
obj["category_id"] = id_map[obj["category_id"]]
objs.append(obj)
record["annotations"] = objs
dataset_dicts.append(record)
if num_instances_without_valid_segmentation > 0:
logger.warning(
"Filtered out {} instances without valid segmentation. "
"There might be issues in your dataset generation process.".
format(num_instances_without_valid_segmentation))
return dataset_dicts
class IterationTimer(HookBase):
"""
Track the time spent for each iteration (each run_step call in the trainer).
Print a summary in the end of training.
This hook uses the time between the call to its :meth:`before_step`
and :meth:`after_step` methods.
Under the convention that :meth:`before_step` of all hooks should only
take negligible amount of time, the :class:`IterationTimer` hook should be
placed at the beginning of the list of hooks to obtain accurate timing.
"""
def __init__(self, warmup_iter=3):
"""
Args:
warmup_iter (int): the number of iterations at the beginning to exclude
from timing.
"""
self._warmup_iter = warmup_iter
self._step_timer = Timer()
def before_train(self):
self._start_time = time.perf_counter()
self._total_timer = Timer()
self._total_timer.pause()
def after_train(self):
logger = logging.getLogger(__name__)
total_time = time.perf_counter() - self._start_time
total_time_minus_hooks = self._total_timer.seconds()
hook_time = total_time - total_time_minus_hooks
num_iter = self.trainer.iter + 1 - self.trainer.start_iter - self._warmup_iter
if num_iter > 0 and total_time_minus_hooks > 0:
# Speed is meaningful only after warmup
# NOTE this format is parsed by grep in some scripts
logger.info(
"Overall training speed: {} iterations in {} ({:.4f} s / it)".
format(
num_iter,
str(datetime.timedelta(
seconds=int(total_time_minus_hooks))),
total_time_minus_hooks / num_iter,
))
logger.info("Total training time: {} ({} on hooks)".format(
str(datetime.timedelta(seconds=int(total_time))),
str(datetime.timedelta(seconds=int(hook_time))),
))
def before_step(self):
self._step_timer.reset()
self._total_timer.resume()
def after_step(self):
# +1 because we're in after_step
iter_done = self.trainer.iter - self.trainer.start_iter + 1
if iter_done >= self._warmup_iter:
sec = self._step_timer.seconds()
self.trainer.storage.put_scalars(time=sec)
else:
self._start_time = time.perf_counter()
self._total_timer.reset()
self._total_timer.pause()
def _load_annotations(self,
json_file,
image_root,
dataset_name=None,
extra_annotation_keys=None):
"""
Load a json file with WiderFace's instances annotation format.
Currently supports instance detection, instance segmentation,
and person keypoints annotations.
Args:
json_file (str): full path to the json file in WiderFace instances annotation format.
image_root (str): the directory where the images in this json file exists.
dataset_name (str): the name of the dataset (e.g., widerface_2019_train).
If provided, this function will also put "thing_classes" into
the metadata associated with this dataset.
extra_annotation_keys (list[str]): list of per-annotation keys that should also be
loaded into the dataset dict (besides "iscrowd", "bbox", "keypoints",
"category_id", "segmentation"). The values for these keys will be returned as-is.
For example, the densepose annotations are loaded in this way.
Returns:
list[dict]: a list of dicts in cvpods standard format. (See
`Using Custom Datasets </tutorials/datasets.html>`_ )
Notes:
1. This function does not read the image files.
The results do not have the "image" field.
"""
from pycocotools.coco import COCO
timer = Timer()
json_file = PathManager.get_local_path(json_file)
with contextlib.redirect_stdout(io.StringIO()):
coco_api = COCO(json_file)
if timer.seconds() > 1:
logger.info("Loading {} takes {:.2f} seconds.".format(
json_file, timer.seconds()))
id_map = None
if dataset_name is not None:
meta = self.meta
cat_ids = sorted(coco_api.getCatIds())
id_map = {v: i for i, v in enumerate(cat_ids)}
meta["thing_dataset_id_to_contiguous_id"] = id_map
# sort indices for reproducible results
img_ids = sorted(list(coco_api.imgs.keys()))
# imgs is a list of dicts, each looks something like:
# {'license': 4,
# 'url': 'http://farm6.staticflickr.com/5454/9413846304_881d5e5c3b_z.jpg',
# 'file_name': 'COCO_val2014_000000001268.jpg',
# 'height': 427,
# 'width': 640,
# 'date_captured': '2013-11-17 05:57:24',
# 'id': 1268}
imgs = coco_api.loadImgs(img_ids)
# anns is a list[list[dict]], where each dict is an annotation
# record for an object. The inner list enumerates the objects in an image
# and the outer list enumerates over images. Example of anns[0]:
# [{'segmentation': [[192.81,
# 247.09,
# ...
# 219.03,
# 249.06]],
# 'area': 1035.749,
# 'iscrowd': 0,
# 'image_id': 1268,
# 'bbox': [192.81, 224.8, 74.73, 33.43],
# 'category_id': 16,
# 'id': 42986},
# ...]
anns = [coco_api.imgToAnns[img_id] for img_id in img_ids]
if "minival" not in json_file:
# The popular valminusminival & minival annotations for COCO2014 contain this bug.
# However the ratio of buggy annotations there is tiny and does not affect accuracy.
# Therefore we explicitly white-list them.
ann_ids = [
ann["id"] for anns_per_image in anns for ann in anns_per_image
]
assert len(set(ann_ids)) == len(ann_ids), \
"Annotation ids in '{}' are not unique!".format(json_file)
imgs_anns = list(zip(imgs, anns))
logger.info("Loaded {} images in COCO format from {}".format(
len(imgs_anns), json_file))
dataset_dicts = []
ann_keys = ["iscrowd", "bbox", "keypoints", "category_id"
] + (extra_annotation_keys or [])
for (img_dict, anno_dict_list) in imgs_anns:
record = {}
record["file_name"] = os.path.join(image_root,
img_dict["file_name"])
record["height"] = img_dict["height"]
record["width"] = img_dict["width"]
image_id = record["image_id"] = img_dict["id"]
objs = []
for anno in anno_dict_list:
# Check that the image_id in this annotation is the same as
# the image_id we're looking at.
# This fails only when the data parsing logic or the annotation file is buggy.
# The original COCO valminusminival2014 & minival2014 annotation files
# actually contains bugs that, together with certain ways of using COCO API,
# can trigger this assertion.
assert anno["image_id"] == image_id
# ensure the width and height of bbox are greater than 0
if anno["bbox"][2] <= 0 or anno["bbox"][3] <= 0:
continue
if anno.get("ignore", 0) != 0:
continue
obj = {key: anno[key] for key in ann_keys if key in anno}
obj["bbox_mode"] = BoxMode.XYWH_ABS
if id_map:
obj["category_id"] = id_map[obj["category_id"]]
objs.append(obj)
record["annotations"] = objs
dataset_dicts.append(record)
return dataset_dicts
def _load_annotations(self):
""" referit dataset have many information:
sentences / tokens / raw
annId -> bbox + mask infor
imgId -> file name + :
"""
timer = Timer()
self.REFER = self._REFER(data_root=osp.join(self.refer_tool_root,
"data/"),
dataset=self.dataset_name,
splitBy=self.dataset_splitby)
dataset_dicts = []
split = self.dataset_split
ref_ids = self.REFER.getRefIds(split=split)
self.proposal_dict = None
if self._proposal_root:
with open(osp.join(self._proposal_root, 'proposals.pkl'),
'rb') as fp:
self.proposal_dict = pickle.load(fp)
for ref_id in tqdm.tqdm(ref_ids):
ref_dict = self.REFER.loadRefs(ref_id)[0]
ann_dict = self.REFER.loadAnns(ref_dict['ann_id'])[0]
img_dict = self.REFER.loadImgs(ref_dict['image_id'])[0]
cat_dict = self.REFER.loadCats(ref_dict['category_id'])[0]
for sentence in ref_dict['sentences']:
dd = {}
dd['sent_id'] = sentence['sent_id']
if self._verbose:
dd['img_dict'] = img_dict
dd['ref_dict'] = ref_dict
dd['cat_dict'] = cat_dict
dd['ann_dict'] = ann_dict
dd['file_name'] = img_dict['file_name']
dd['image_size'] = [img_dict['height'], img_dict['width']]
dd['tokens'] = sentence['tokens']
dd['raw'] = sentence['raw']
dd['category_id'] = None # cat_dict[] # FIXME (增加cat的逻辑)
dd['bbox'] = self.REFER.getRefBox(ref_id)
if len(dd['bbox']) != 4:
print('invalid bbox')
continue
dd['height'] = img_dict['height']
dd['width'] = img_dict['width']
# FIXME (增加gt_classes信息)
b = dd['bbox']
dataset_dicts.append(dd)
logging.info("Loading {} takes {:.2f} seconds.".format(
self.raw_name, timer.seconds()))
dataset_tranforms = [
#DatasetTransform_Unknowlize_Truncate(False, -1, 15, ['tokens'], ['token_ids']),
#DatasetTransform_Tokenize(False, True, ['token_ids'], ['token_ids'], self._is_train, osp.join(self._output_dir, "vocabulary.pkl")),
#DatasetTransform_NegativeTokensSample(1, ['token_ids'], ['neg_token_ids'])
]
if not self._image_only:
dataset_tranforms.append(DatasetTransform_ExcludeEmptyProposal())
#{{{ add this logic to judge : whether small gtboxes will cause no proper proposals
self._exclude_small_gtboxes = not self._is_train and not self._image_only
if self._exclude_small_gtboxes:
dataset_tranforms.append(DatasetTransform_ExcludeSmallGt(20000)),
#}}}
if self._add_attribute:
atts_json_file = osp.join(
self.refer_tool_root, "data/", "parsed_atts",
self.dataset_name + '_' + self.dataset_splitby, "sents.json")
dataset_tranforms.append(
DatasetTransform_AddAttribute(
sent_json_filepath=atts_json_file))
dataset_dicts, context = self._start_dataset_transform(
dataset_dicts, dataset_tranforms,
{'proposal_dict': self.proposal_dict})
self.context = context
if not self._is_train and not self._image_only: #{{{
#indices = np.random.randint(0, len(dataset_dicts), 1000)
#self.dataset_dicts = [dataset_dicts[i] for i in indices]
self.dataset_dicts = dataset_dicts
else:
self.dataset_dicts = dataset_dicts #}}}
# if image_only ： 那么将所有相同名字的过滤掉
if self._image_only:
print('dataset_len:', len(dataset_dicts))
print('Image Only Mode!!')
image_pool = {}
new_dict = []
for item in self.dataset_dicts:
if item['file_name'] not in image_pool:
new_dict.append(item)
image_pool[item['file_name']] = 1
print('dataset_len:', len(new_dict))