# ------------------------------------------------------------------------------
# Copyright (c) Microsoft
# Licensed under the MIT License.
# Written by Bowen Cheng (bcheng9@illinois.edu) and Bin Xiao (leoxiaobin@gmail.com)
# ------------------------------------------------------------------------------
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from collections import defaultdict
from collections import OrderedDict
import logging
import os
import os.path
import cv2
import json_tricks as json
import numpy as np
from torch.utils.data import Dataset
from crowdposetools.cocoeval import COCOeval
# from dataset.JointsDataset import JointsDataset
from dataset.JointsDataset import JointsDataset
from nms.nms import oks_nms
from nms.nms import soft_oks_nms
# -------------------------------------------
crowdpose_sigmas = np.array([.79, .79, .72, .72, .62, .62, 1.07, 1.07, .87, .87, .89, .89, .25, .25]) / 10.0
# -------------------------------------------
logger = logging.getLogger(__name__)
class EXLPOSEDataset(JointsDataset):
"""`CrowdPose`_ Dataset.
Args:
root (string): Root directory where dataset is located to.
dataset (string): Dataset name(train2017, val2017, test2017).
data_format(string): Data format for reading('jpg', 'zip')
transform (callable, optional): A function/transform that takes in an opencv image
and returns a transformed version. E.g, ``transforms.ToTensor``
target_transform (callable, optional): A function/transform that takes in the
target and transforms it.
"""
def __init__(self, cfg, root, image_set, is_train, transform=None):
super().__init__(cfg, root, image_set, is_train, transform=None)
# def __init__(self, root, dataset, data_format, transform=None,
# target_transform=None):
from crowdposetools.coco import COCO
self.nms_thre = cfg.TEST.NMS_THRE
self.image_thre = cfg.TEST.IMAGE_THRE
self.soft_nms = cfg.TEST.SOFT_NMS
self.oks_thre = cfg.TEST.OKS_THRE
self.in_vis_thre = cfg.TEST.IN_VIS_THRE
self.bbox_file = cfg.TEST.COCO_BBOX_FILE
self.use_gt_bbox = cfg.TEST.USE_GT_BBOX
self.image_width = cfg.MODEL.IMAGE_SIZE[0]
self.image_height = cfg.MODEL.IMAGE_SIZE[1]
self.aspect_ratio = self.image_width * 1.0 / self.image_height
self.pixel_std = 200
self.root = cfg.DATASET.ROOT
# self.scale_thre = cfg.TEST.SCALE_THRE
# self.dataset_type = dataset_type
# self.coco_WL = COCO(self._get_anno_file_keypoint()[0])
# self.coco_LL = COCO(self._get_anno_file_keypoint()[1])
self.coco = COCO(self._get_anno_file_keypoint())
# self.ids_WL = list(self.coco_WL.imgs.keys())
self.ids = list(self.coco.imgs.keys())
self.transform = transform
cats = [cat['name']
for cat in self.coco.loadCats(self.coco.getCatIds())]
self.classes = ['__background__'] + cats
logger.info('=> classes: {}'.format(self.classes))
self.num_classes = len(self.classes)
self._class_to_ind = dict(zip(self.classes, range(self.num_classes)))
self._class_to_coco_ind = dict(zip(cats, self.coco.getCatIds()))
self._coco_ind_to_class_ind = dict(
[
(self._class_to_coco_ind[cls], self._class_to_ind[cls])
for cls in self.classes[1:]
]
)
# load image file names
self.image_set_index = self._load_image_set_index()
self.num_images = len(self.image_set_index)
logger.info('=> num_images: {}'.format(self.num_images))
self.num_joints = 14
self.flip_pairs = [[0, 1], [2, 3], [4, 5], [6, 7],
[8, 9], [10, 11]]
self.parent_ids = None
self.upper_body_ids = (0, 1, 2, 3, 4, 5, 12, 13)
self.lower_body_ids = (6, 7, 8, 9, 10, 11)
self.joints_weight = np.array(
[
1., 1., 1.2, 1.2,
1.5, 1.5, 1., 1.,
1.2, 1.2, 1.5, 1.5,
1., 1.
],
dtype=np.float32
).reshape((self.num_joints, 1))
self.db = self._get_db()
if is_train and cfg.DATASET.SELECT_DATA:
self.db = self.select_data(self.db)
logger.info('=> load {} samples'.format(len(self.db)))
return
def _load_image_set_index(self):
""" image id: int """
image_ids = self.coco.getImgIds()
return image_ids
def _get_anno_file_keypoint(self):
""" self.root / annotations / person_keypoints_train2017.json """
prefix = 'ExLPose' \
if 'test' not in self.image_set else 'ExLPose'
return os.path.join(
self.root,
'Annotations',
prefix + '_' + self.image_set + '.json'
)
def _get_db(self):
if self.is_train or self.use_gt_bbox:
# use ground truth bbox
gt_db = self._load_coco_keypoint_annotations()
else:
# use bbox from detection
gt_db = self._load_coco_person_detection_results()
return gt_db
def _load_coco_keypoint_annotations(self):
""" ground truth bbox and keypoints """
gt_db = []
for index in self.image_set_index:
gt_db.extend(self._load_coco_keypoint_annotation_kernal(index))
return gt_db
# def _get_image_path(self, file_name):
# # return os.path.join(self.image_dir, file_name)
# images_dir = os.path.join(self.root, 'ExLPose')
# return os.path.join(images_dir, file_name)
def image_path_from_index(self, index):
""" example: images / train2017 / 000000119993.jpg """
# print('index',index)
# file_name = '%012d.jpg' % index
# print('file_name', file_name)
# load well-light image
# coco_WL = self.coco_WL
# img_id_WL = self.ids_WL[index]
# ann_ids_WL = coco_WL.getAnnIds(imgIds=img_id_WL)
# target_WL = coco_WL.loadAnns(ann_ids_WL)
# image_info_WL = coco_WL.loadImgs(img_id_WL)[0]
# file_name_WL = image_info_WL['file_name']
# load unpaired low-lit image during training
coco_LL = self.coco
index_LL = np.random.randint(len(self.ids))
img_id_LL = self.ids[index_LL]
ann_ids_LL = coco_LL.getAnnIds(imgIds=img_id_LL)
# target_LL = coco_LL.loadAnns(ann_ids_LL)
image_info_LL = coco_LL.loadImgs(img_id_LL)[0]
file_name_LL = image_info_LL['file_name']
# image_path_wl = os.path.join(self.root, file_name_WL)
image_path_ll = os.path.join(self.root, file_name_LL)
return image_path_ll
def _load_coco_keypoint_annotation_kernal(self, index):
"""
coco ann: [u'segmentation', u'area', u'iscrowd', u'image_id', u'bbox', u'category_id', u'id']
iscrowd:
crowd instances are handled by marking their overlaps with all categories to -1
and later excluded in training
bbox:
[x1, y1, w, h]
:param index: coco image id
:return: db entry
"""
# im_ann_wl = self.coco.loadImgs(index)[0]
im_ann_ll = self.coco.loadImgs(index)
# print :im_ann_ll:im_ann_ll[0]
# print('im_ann_ll[0]',im_ann_ll[0])
#im_ann_ll[0] {'file_name': 'dark/imgs_0209_vid000011_exp1200_dark_000052__gain_5.60_exposure_333.png', 'id': 2527, 'height': 1199, 'width': 1919, 'crowdIndex': 0}
#print :im_ann_ll
# print('im_ann_ll',im_ann_ll)
#im_ann_ll [{'file_name': 'dark/imgs_0209_vid000011_exp1200_dark_000052__gain_5.60_exposure_333.png', 'id': 2527, 'height': 1199, 'width': 1919, 'crowdIndex': 0}]
width = im_ann_ll[0]['width']
height = im_ann_ll[0]['height']
# image_id=im_ann_ll[0]['image_id']
annIds = self.coco.getAnnIds(imgIds=index, iscrowd=False)
objs = self.coco.loadAnns(annIds)
# print('objs',objs)
# objs[{'num_keypoints': 13, 'iscrowd': 0,
# 'keypoints': [1171.24, 545.66, 2, 1195.69, 545.66, 2, 0, 0, 0, 1236.45, 586.41, 2, 1255.47, 564.68, 2,
# 1277.2, 567.39, 2, 1192.98, 635.32, 2, 1214.71, 640.75, 2, 1214.71, 733.13, 2, 1222.86,
# 733.13, 2, 1217.43, 806.49, 2, 1217.43, 809.2, 2, 1192.98, 491.32, 2, 1190.26, 532.07, 2],
# 'image_id': 675, 'bbox': [1157.04, 487.24, 117.05999999999995, 346.77], 'category_id': 1, 'id': 934}, {
# 'num_keypoints': 14, 'iscrowd': 0,
# 'keypoints': [1013.08, 524.59, 2, 1068.17, 518.79, 2, 1004.38, 591.28, 2, 1091.36, 565.18, 2, 1004.38,
# 649.27, 2, 1091.36, 608.68, 2, 1027.57, 620.28, 2, 1065.27, 620.28, 2, 1033.37, 707.26,
# 2, 1071.07, 704.36, 2, 1039.17, 785.55, 2, 1065.27, 782.65, 2, 1050.77, 455.0, 2,
# 1044.97, 501.39, 2], 'image_id': 675,
# 'bbox': [996.75, 447.75, 92.54999999999995, 370.15], 'category_id': 1, 'id': 935}, {
# 'num_keypoints': 14, 'iscrowd': 0,
# 'keypoints': [938.28, 632.51, 2, 990.4, 632.51, 2, 929.6, 672.47, 2, 1004.3, 674.21, 2, 933.07, 702.0,
# 2, 1000.83, 696.79, 2, 950.44, 714.17, 2, 985.19, 714.17, 2, 957.39, 757.6, 2, 983.45,
# 757.6, 2, 960.87, 794.08, 2, 979.98, 794.08, 2, 964.34, 576.92, 2, 964.34, 616.88, 2],
# 'image_id': 675, 'bbox': [931.86, 586.48, 76.12, 221.37], 'category_id': 1, 'id': 936}, {
# 'num_keypoints': 6, 'iscrowd': 0,
# 'keypoints': [0, 0, 0, 310.4, 581.7, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 319.1, 683.5, 2, 354.8,
# 682.1, 2, 0, 0, 0, 338.9, 754.5, 2, 0, 0, 0, 343.2, 831.3, 2, 0, 0, 0, 286.8, 574.0, 2],
# 'image_id': 675, 'bbox': [278.3, 560.83, 91.75, 294.74], 'category_id': 1, 'id': 937}, {
# 'num_keypoints': 14, 'iscrowd': 0,
# 'keypoints': [800.73, 497.26, 2, 704.74, 504.94, 2, 831.45, 562.54, 2, 681.7, 570.22, 2, 816.09,
# 589.42, 2, 735.46, 585.58, 2, 796.89, 643.17, 2, 735.46, 647.01, 2, 796.89, 735.33, 2,
# 743.13, 743.01, 2, 777.69, 808.28, 2, 758.49, 835.16, 2, 750.81, 408.95, 2, 750.81,
# 478.06, 2], 'image_id': 675, 'bbox': [668.02, 399.45, 172.63, 477.93], 'category_id': 1,
# 'id': 938}, {'num_keypoints': 12, 'iscrowd': 0,
# 'keypoints': [384.14, 555.52, 2, 459.02, 558.52, 2, 375.15, 609.44, 2, 465.01, 612.44, 2,
# 0, 0, 0, 0, 0, 0, 402.11, 669.35, 2, 444.05, 663.36, 2, 399.12, 741.23, 2,
# 441.05, 738.24, 2, 405.11, 819.11, 2, 435.06, 816.12, 2, 414.09, 486.63, 2,
# 420.08, 528.57, 2], 'image_id': 675,
# 'bbox': [350.28, 476.15, 123.63000000000005, 382.4], 'category_id': 1, 'id': 939}, {
# 'num_keypoints': 12, 'iscrowd': 0,
# 'keypoints': [1081.23, 550.46, 2, 1124.05, 550.46, 2, 0, 0, 0, 1161.15, 581.86, 2, 0, 0, 0, 1149.74,
# 541.9, 2, 1089.79, 647.52, 2, 1129.76, 650.37, 2, 1115.48, 727.44, 2, 1135.46, 730.3, 2,
# 1122.2, 814.1, 2, 1130.9, 811.8, 2, 1109.77, 487.67, 2, 1106.92, 533.34, 2],
# 'image_id': 675, 'bbox': [1062.76, 483.38, 118.72000000000003, 364.38], 'category_id': 1, 'id': 940}, {
# 'num_keypoints': 14, 'iscrowd': 0,
# 'keypoints': [927.8, 528.25, 2, 843.73, 525.01, 2, 950.44, 589.68, 2, 814.63, 579.98, 2, 934.27,
# 644.66, 2, 827.56, 647.89, 2, 914.87, 651.12, 2, 856.67, 651.12, 2, 914.87, 735.19, 2,
# 885.77, 728.73, 2, 901.94, 809.57, 2, 895.47, 825.73, 2, 876.07, 453.88, 2, 882.53,
# 512.08, 2], 'image_id': 675, 'bbox': [821.0, 449.03, 144.70000000000005, 412.89],
# 'category_id': 1, 'id': 941}, {'num_keypoints': 5, 'iscrowd': 0,
# 'keypoints': [492.7, 521.0, 2, 480.16, 510.64, 2, 495.5, 587.2, 2, 0, 0,
# 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
# 0, 0, 0, 0, 0, 0, 431.6, 470.7, 2, 467.6, 506.2, 2],
# 'image_id': 675,
# 'bbox': [422.0, 462.48, 95.29999999999995, 246.51999999999998],
# 'category_id': 1, 'id': 942}]
# sanitize bboxes
valid_objs = []
for obj in objs:
x, y, w, h = obj['bbox']
x1 = np.max((0, x))
y1 = np.max((0, y))
x2 = np.min((width - 1, x1 + np.max((0, w - 1))))
y2 = np.min((height - 1, y1 + np.max((0, h - 1))))
if x2 >= x1 and y2 >= y1:
obj['clean_bbox'] = [x1, y1, x2 - x1, y2 - y1]
valid_objs.append(obj)
objs = valid_objs
rec = []
for obj in objs:
image_id=obj['image_id']
print('image_id',image_id)#image_id 0,image_id 2
cls = self._coco_ind_to_class_ind[obj['category_id']]
if cls != 1:
continue
# ignore objs without keypoints annotation
if max(obj['keypoints']) == 0:
continue
joints_3d = np.zeros((self.num_joints, 3), dtype=np.float)
joints_3d_vis = np.zeros((self.num_joints, 3), dtype=np.float)
for ipt in range(self.num_joints):
joints_3d[ipt, 0] = obj['keypoints'][ipt * 3 + 0]
joints_3d[ipt, 1] = obj['keypoints'][ipt * 3 + 1]
joints_3d[ipt, 2] = 0
t_vis = obj['keypoints'][ipt * 3 + 2]
if t_vis > 1:
t_vis = 1
joints_3d_vis[ipt, 0] = t_vis
joints_3d_vis[ipt, 1] = t_vis
joints_3d_vis[ipt, 2] = 0
center, scale = self._box2cs(obj['clean_bbox'][:4])
# image_file_name = im_ann_ll['file_name'].split('/')[-1]
# image_path = os.path.join(self.image_dir, image_file_name)
print('self.image_path_from_index(index)',self.image_path_from_index(index))
rec.append({
'image': self.image_path_from_index(index),
'image_id': image_id,
'center': center,
'scale': scale,
'joints_3d': joints_3d,
'joints_3d_vis': joints_3d_vis,
'filename': '',
'imgnum': 0,
# 'annotation_id': obj['id']
})
return rec
def _box2cs(self, box):
x, y, w, h = box[:4]
return self._xywh2cs(x, y, w, h)
def _xywh2cs(self, x, y, w, h):
center = np.zeros((2), dtype=np.float32)
center[0] = x + w * 0.5
center[1] = y + h * 0.5
if w > self.aspect_ratio * h:
h = w * 1.0 / self.aspect_ratio
elif w < self.aspect_ratio * h:
w = h * self.aspect_ratio
scale = np.array(
[w * 1.0 / self.pixel_std, h * 1.0 / self.pixel_std],
dtype=np.float32)
if center[0] != -1:
scale = scale * 1.25
# scale = scale * self.scale_thre
return center, scale
def _load_coco_person_detection_results(self):
all_boxes = None
with open(self.bbox_file, 'r') as f:
all_boxes = json.load(f)
if not all_boxes:
logger.error('=> Load %s fail!' % self.bbox_file)
return None
logger.info('=> Total boxes: {}'.format(len(all_boxes)))
kpt_db = []
num_boxes = 0
for n_img in range(0, len(all_boxes)):
det_res = all_boxes[n_img]
if det_res['category_id'] != 1:
continue
img_name = self.image_path_from_index(det_res['image_id'])
image_id = det_res['image_id']
box = det_res['bbox']
score = det_res['score']
if score < self.image_thre:
continue
num_boxes = num_boxes + 1
center, scale = self._box2cs(box)
joints_3d = np.zeros((self.num_joints, 3), dtype=np.float)
joints_3d_vis = np.ones(
(self.num_joints, 3), dtype=np.float)
kpt_db.append({
'image': img_name,
'image_id': image_id,
'center': center,
'scale': scale,
'score': score,
'joints_3d': joints_3d,
'joints_3d_vis': joints_3d_vis,
})
logger.info('=> Total boxes after filter low score@{}: {}'.format(
self.image_thre, num_boxes))
return kpt_db
def __repr__(self):
fmt_str = 'Dataset ' + self.__class__.__name__ + '\n'
fmt_str += ' Number of datapoints: {}\n'.format(self.__len__())
fmt_str += ' Root Location: {}\n'.format(self.root)
# tmp = ' Transforms (if any): '
# fmt_str += '{0}{1}\n'.format(tmp, self.transform.__repr__().replace('\n', '\n' + ' ' * len(tmp)))
# tmp = ' Target Transforms (if any): '
# fmt_str += '{0}{1}'.format(tmp, self.target_transform.__repr__().replace('\n', '\n' + ' ' * len(tmp)))
return fmt_str
def processKeypoints(self, keypoints):
tmp = keypoints.copy()
if keypoints[:, 2].max() > 0:
p = keypoints[keypoints[:, 2] > 0][:, :2].mean(axis=0)
num_keypoints = keypoints.shape[0]
for i in range(num_keypoints):
tmp[i][0:3] = [
float(keypoints[i][0]),
float(keypoints[i][1]),
float(keypoints[i][2])
]
return tmp
# def evaluate(self, cfg, preds, scores, output_dir,
# *args, **kwargs):
def evaluate(self, cfg, preds, output_dir, all_boxes, image_id,
*args, **kwargs):
rank = cfg.RANK
# if all_boxes.shape[1] == 8:
# return self.evaluate_lambda(cfg, preds, output_dir, all_boxes, img_path, epoch, *args, **kwargs)
# print('img_path', img_path)#'/media/zou/D/dataset/ExLPose/dark/imgs_0206_vid000012_exp400_dark_000052__gain_4.10_exposure_1250.png'
res_folder = os.path.join(output_dir, 'results')
if not os.path.exists(res_folder):
try:
os.makedirs(res_folder)
except Exception:
logger.error('Fail to make {}'.format(res_folder))
res_file = os.path.join(
res_folder, 'keypoints_{}_results_{}.json'.format(
exlpose
于 2024-12-15 13:55:32 首次发布
最低0.47元/天 解锁文章
扫一扫
2792
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
