def __init__(self):
ImageRoot = './data/coco2017/train2017'
AnnoFile = './data/coco2017/annotations/person_keypoints_train2017_pose2seg.json'
self.datainfos = CocoDatasetInfo(ImageRoot,
AnnoFile,
onlyperson=True,
loadimg=True)
def __init__(self, batch_size, shuffle, num_workers):
super().__init__(batch_size=batch_size,
shuffle=shuffle,
num_workers=num_workers)
ImageRoot = './data/coco2017/train2017'
AnnoFile = './data/coco2017/annotations/person_keypoints_train2017_pose2seg.json'
self.datainfos = CocoDatasetInfo(ImageRoot,
AnnoFile,
onlyperson=True,
loadimg=True)
self.total_len = len(self.datainfos)
def test(model, dataset='cocoVal', logger=print):
ImageRoot = './data/coco2017/val2017'
AnnoFile = './data/person_keypoints_val2017_pose2seg.json'
datainfos = CocoDatasetInfo(ImageRoot, AnnoFile, onlyperson=True, loadimg=True)
model.eval()
results_segm = []
imgIds = []
for i in tqdm(range(len(datainfos))):
rawdata = datainfos[i]
img = rawdata['data']
image_id = rawdata['id']
height, width = img.shape[0:2]
gt_kpts = np.float32(rawdata['gt_keypoints']).transpose(0, 2, 1) # (N, 17, 3)
gt_segms = rawdata['segms']
gt_masks = np.array([annToMask(segm, height, width) for segm in gt_segms])
output = model([img], [gt_kpts], [gt_masks])
for mask in output[0]:
maskencode = maskUtils.encode(np.asfortranarray(mask))
maskencode['counts'] = maskencode['counts'].decode('ascii')
results_segm.append({
"image_id": image_id,
"category_id": 1,
"score": 1.0,
"segmentation": maskencode
})
imgIds.append(image_id)
def do_eval_coco(image_ids, coco, results, flag):
from pycocotools.cocoeval import COCOeval
assert flag in ['bbox', 'segm', 'keypoints']
# Evaluate
coco_results = coco.loadRes(results)
cocoEval = COCOeval(coco, coco_results, flag)
cocoEval.params.imgIds = image_ids
cocoEval.params.catIds = [1]
cocoEval.evaluate()
cocoEval.accumulate()
cocoEval.summarize()
return cocoEval
cocoEval = do_eval_coco(imgIds, datainfos.COCO, results_segm, 'segm')
logger('[POSE2SEG] AP|.5|.75| S| M| L| AR|.5|.75| S| M| L|')
_str = '[segm_score] %s '%dataset
for value in cocoEval.stats.tolist():
_str += '%.3f '%value
logger(_str)
cocoEval.accumulate()
cocoEval.summarize()
return cocoEval
cocoEval = do_eval_coco(imgIds, datainfos.COCO, results_segm, 'segm')
logger('[POSE2SEG] AP|.5|.75| S| M| L| AR|.5|.75| S| M| L|')
_str = '[segm_score] %s '%dataset
for value in cocoEval.stats.tolist():
_str += '%.3f '%value
logger(_str)
if __name__=='__main__':
model_path = './pose2seg_release.pkl'
print('===========> loading model <===========')
model = Pose2Seg().cuda()
model.init(model_path)
model.eval()
ImageRoot = './data/images'
AnnoFile = './data/person_keypoints_val2017_pose2seg.json'
datainfos = CocoDatasetInfo(ImageRoot, AnnoFile, onlyperson=True, loadimg=True)
model.eval()
results_segm = []
imgIds = []
for i in tqdm(range(len(datainfos))):
rawdata = datainfos[i]
g1 = np.float32(rawdata['gt_keypoints'])
g2 = np.float32(rawdata['gt_keypoints']).transpose(0, 2, 1)
def cluster(dataset='coco', cat_num=3, vis_threshold=0.4,
minpoints=8, save_file='./modeling/templates2.json', visualize=False):
# We try `cat_num` from 1 to 6 multiple times. we want to see
# what the cluster centers look like when vary the numbers of
# group. While the kmean method, which is heavily relay on the
# initial status, gives nearly the same cluster centers when
# `cat_num` = 3 each time. So we assume the coco dataset accurately
# have 3 clusters.(a TODO is to visualize this dataset.) And
# the visualization of the cluster centers seems to reasonable:
# (1) a full body. (2) a full body without head (3) an upper body.
# Note that (2) seems representing the backward of a person.
if dataset == 'coco':
datainfos = CocoDatasetInfo('./data/coco2017/train2017',
'./data/coco2017/annotations/person_keypoints_train2017_pose2seg.json',
loadimg=False)
connections = [[16, 14], [14, 12], [17, 15], [15, 13], [12, 13],
[6, 12], [7, 13], [6, 7], [6, 8], [7, 9], [8, 10],
[9, 11], [2, 3], [1, 2], [1, 3], [2, 4], [3, 5], [4, 6], [5, 7]]
names = ["nose",
"left_eye", "right_eye",
"left_ear", "right_ear",
"left_shoulder", "right_shoulder",
"left_elbow", "right_elbow",
"left_wrist", "right_wrist",
"left_hip", "right_hip",
"left_knee", "right_knee",
"left_ankle", "right_ankle"]
flip_map = {'left_eye': 'right_eye',
'left_ear': 'right_ear',
'left_shoulder': 'right_shoulder',
'left_elbow': 'right_elbow',
'left_wrist': 'right_wrist',
'left_hip': 'right_hip',
'left_knee': 'right_knee',
'left_ankle': 'right_ankle'}
def flip_keypoints(keypoints, keypoint_flip_map, keypoint_coords, width):
"""Left/right flip keypoint_coords. keypoints and keypoint_flip_map are
accessible from get_keypoints().
"""
flipped_kps = keypoint_coords.copy()
for lkp, rkp in keypoint_flip_map.items():
lid = keypoints.index(lkp)
rid = keypoints.index(rkp)
flipped_kps[:, :, lid] = keypoint_coords[:, :, rid]
flipped_kps[:, :, rid] = keypoint_coords[:, :, lid]
# Flip x coordinates
flipped_kps[:, 0, :] = width - flipped_kps[:, 0, :]
# Maintain COCO convention that if visibility == 0, then x, y = 0
inds = np.where(flipped_kps[:, 2, :] == 0)
flipped_kps[inds[0], 0, inds[1]] = 0
return flipped_kps
all_kpts = []
for idx in tqdm(range(len(datainfos))):
rawdata = datainfos[idx]
gt_boxes = rawdata['boxes']
gt_kpts = rawdata['gt_keypoints'].transpose(0, 2, 1) # (N, 17, 3)
gt_ignores = rawdata['is_crowd']
normed_kpts = norm_kpt_by_box(gt_kpts, gt_boxes)
normed_kpts_flipped = flip_keypoints(names, flip_map,
normed_kpts.transpose(0, 2, 1), 1.0).transpose(0, 2, 1)
normed_kpts = np.vstack((normed_kpts, normed_kpts_flipped))
for kpt in normed_kpts:
if np.sum(kpt) == 0:
continue
elif np.sum(kpt[:, 2] > 0) < minpoints:
continue
else:
all_kpts.append(kpt)
all_kpts = np.array(all_kpts)
print('data to be clustered:', all_kpts.shape)
res = cluster_zixi(all_kpts, cat_num)
save_dict = {}
save_dict['connections'] = connections
save_dict['names'] = names
save_dict['flip_map'] = flip_map
save_dict['vis_threshold'] = vis_threshold
save_dict['minpoints'] = minpoints
save_dict['templates'] = [item.tolist() for item in res[0]]
if save_file is not None:
with open(save_file, 'w') as result_file:
json.dump(save_dict, result_file)
if visualize:
for center in res[0]:
center = center.reshape(-1, 3)
draw_skeleton(center, 200, 200, vis_threshold)
print('cluster() done.')
return res
else:
raise NotImplementedError
def test(model, dataset='cocoVal', logger=print):
if dataset == 'OCHumanVal':
ImageRoot = r'\\fs01\Algo\ML\Datasets\Pose2Seg\data\OCHuman\images'
AnnoFile = r'\\fs01\Algo\ML\Datasets\Pose2Seg\data\OCHuman\annotations\ochuman_coco_format_val_range_0.00_1.00.json'
elif dataset == 'OCHumanTest':
ImageRoot = r'\\fs01\Algo\ML\Datasets\Pose2Seg\data\OCHuman\images'
AnnoFile = r'\\fs01\Algo\ML\Datasets\Pose2Seg\data\OCHuman\annotations\ochuman_coco_format_test_range_0.00_1.00.json'
elif dataset == 'cocoVal':
ImageRoot = r'\\fs01\Algo\ML\Datasets\Pose2Seg\data\coco2017\val2017'
AnnoFile = r'\\fs01\Algo\ML\Datasets\Pose2Seg\data\coco2017\annotations\person_keypoints_val2017_pose2seg.json'
datainfos = CocoDatasetInfo(ImageRoot,
AnnoFile,
onlyperson=True,
loadimg=True)
model.eval()
results_segm = []
imgIds = []
for i in tqdm(range(3, len(datainfos))):
rawdata = datainfos[i]
img = rawdata['data']
image_id = rawdata['id']
height, width = img.shape[0:2]
gt_kpts = np.float32(rawdata['gt_keypoints']).transpose(
0, 2, 1) # (N, 17, 3)
gt_segms = rawdata['segms']
gt_masks = np.array(
[annToMask(segm, height, width) for segm in gt_segms])
output = model([img], [gt_kpts], [gt_masks])
img = img[..., ::-1]
# plt.switch_backend("TkAgg")
# plt.imshow(gt_masks[0,...])
# plt.show()
# fig.savefig('C:\\Users\\erez\\Projects\\Pose2Seg\\demo.png', bbox_inches='tight')
MASKS = np.zeros(output[0][0].shape)
for id, mask in enumerate(output[0]):
from skimage.color import label2rgb
MASKS += (id + 1) * mask
maskencode = maskUtils.encode(np.asfortranarray(mask))
maskencode['counts'] = maskencode['counts'].decode('ascii')
results_segm.append({
"image_id": image_id,
"category_id": 1,
"score": 1.0,
"segmentation": maskencode
})
imgIds.append(image_id)
image_label_overlay = label2rgb(MASKS,
image=img,
alpha=0.3,
bg_label=0)
plt.imshow(image_label_overlay)
plt.show()
def do_eval_coco(image_ids, coco, results, flag):
from pycocotools.cocoeval import COCOeval
assert flag in ['bbox', 'segm', 'keypoints']
# Evaluate
coco_results = coco.loadRes(results)
cocoEval = COCOeval(coco, coco_results, flag)
cocoEval.params.imgIds = image_ids
cocoEval.params.catIds = [1]
cocoEval.evaluate()
cocoEval.accumulate()
cocoEval.summarize()
return cocoEval
cocoEval = do_eval_coco(imgIds, datainfos.COCO, results_segm, 'segm')
logger('[POSE2SEG] AP|.5|.75| S| M| L| AR|.5|.75| S| M| L|')
_str = '[segm_score] %s ' % dataset
for value in cocoEval.stats.tolist():
_str += '%.3f ' % value
logger(_str)
def __init__(self):
ImageRoot = r'\\fs01\Algo\ML\Datasets\Pose2Seg\data\coco2017\train2017'
AnnoFile = r'\\fs01\Algo\ML\Datasets\Pose2Seg\data\coco2017\annotations\person_keypoints_train2017_pose2seg.json'
self.datainfos = CocoDatasetInfo(ImageRoot, AnnoFile, onlyperson=True, loadimg=True)