485 lines
15 KiB
Python
485 lines
15 KiB
Python
import copy
|
|
import json
|
|
import os
|
|
from typing import List
|
|
|
|
import cv2
|
|
import matplotlib
|
|
import numpy as np
|
|
|
|
import draw_utils
|
|
import triangulate_poses
|
|
import utils_2d_pose
|
|
from skelda import utils_pose
|
|
|
|
# ==================================================================================================
|
|
|
|
filepath = os.path.dirname(os.path.realpath(__file__)) + "/"
|
|
test_img_dir = filepath + "../data/"
|
|
whole_body = {
|
|
"foots": False,
|
|
"face": False,
|
|
"hands": False,
|
|
}
|
|
|
|
joint_names_2d = [
|
|
"nose",
|
|
"eye_left",
|
|
"eye_right",
|
|
"ear_left",
|
|
"ear_right",
|
|
"shoulder_left",
|
|
"shoulder_right",
|
|
"elbow_left",
|
|
"elbow_right",
|
|
"wrist_left",
|
|
"wrist_right",
|
|
"hip_left",
|
|
"hip_right",
|
|
"knee_left",
|
|
"knee_right",
|
|
"ankle_left",
|
|
"ankle_right",
|
|
]
|
|
if whole_body["foots"]:
|
|
joint_names_2d.extend(
|
|
[
|
|
"foot_toe_big_left",
|
|
"foot_toe_small_left",
|
|
"foot_heel_left",
|
|
"foot_toe_big_right",
|
|
"foot_toe_small_right",
|
|
"foot_heel_right",
|
|
]
|
|
)
|
|
if whole_body["face"]:
|
|
joint_names_2d.extend(
|
|
[
|
|
"face_jaw_right_1",
|
|
"face_jaw_right_2",
|
|
"face_jaw_right_3",
|
|
"face_jaw_right_4",
|
|
"face_jaw_right_5",
|
|
"face_jaw_right_6",
|
|
"face_jaw_right_7",
|
|
"face_jaw_right_8",
|
|
"face_jaw_middle",
|
|
"face_jaw_left_1",
|
|
"face_jaw_left_2",
|
|
"face_jaw_left_3",
|
|
"face_jaw_left_4",
|
|
"face_jaw_left_5",
|
|
"face_jaw_left_6",
|
|
"face_jaw_left_7",
|
|
"face_jaw_left_8",
|
|
"face_eyebrow_right_1",
|
|
"face_eyebrow_right_2",
|
|
"face_eyebrow_right_3",
|
|
"face_eyebrow_right_4",
|
|
"face_eyebrow_right_5",
|
|
"face_eyebrow_left_1",
|
|
"face_eyebrow_left_2",
|
|
"face_eyebrow_left_3",
|
|
"face_eyebrow_left_4",
|
|
"face_eyebrow_left_5",
|
|
"face_nose_1",
|
|
"face_nose_2",
|
|
"face_nose_3",
|
|
"face_nose_4",
|
|
"face_nose_5",
|
|
"face_nose_6",
|
|
"face_nose_7",
|
|
"face_nose_8",
|
|
"face_nose_9",
|
|
"face_eye_right_1",
|
|
"face_eye_right_2",
|
|
"face_eye_right_3",
|
|
"face_eye_right_4",
|
|
"face_eye_right_5",
|
|
"face_eye_right_6",
|
|
"face_eye_left_1",
|
|
"face_eye_left_2",
|
|
"face_eye_left_3",
|
|
"face_eye_left_4",
|
|
"face_eye_left_5",
|
|
"face_eye_left_6",
|
|
"face_mouth_1",
|
|
"face_mouth_2",
|
|
"face_mouth_3",
|
|
"face_mouth_4",
|
|
"face_mouth_5",
|
|
"face_mouth_6",
|
|
"face_mouth_7",
|
|
"face_mouth_8",
|
|
"face_mouth_9",
|
|
"face_mouth_10",
|
|
"face_mouth_11",
|
|
"face_mouth_12",
|
|
"face_mouth_13",
|
|
"face_mouth_14",
|
|
"face_mouth_15",
|
|
"face_mouth_16",
|
|
"face_mouth_17",
|
|
"face_mouth_18",
|
|
"face_mouth_19",
|
|
"face_mouth_20",
|
|
]
|
|
)
|
|
if whole_body["hands"]:
|
|
joint_names_2d.extend(
|
|
[
|
|
"hand_wrist_left",
|
|
"hand_finger_thumb_left_1",
|
|
"hand_finger_thumb_left_2",
|
|
"hand_finger_thumb_left_3",
|
|
"hand_finger_thumb_left_4",
|
|
"hand_finger_index_left_1",
|
|
"hand_finger_index_left_2",
|
|
"hand_finger_index_left_3",
|
|
"hand_finger_index_left_4",
|
|
"hand_finger_middle_left_1",
|
|
"hand_finger_middle_left_2",
|
|
"hand_finger_middle_left_3",
|
|
"hand_finger_middle_left_4",
|
|
"hand_finger_ring_left_1",
|
|
"hand_finger_ring_left_2",
|
|
"hand_finger_ring_left_3",
|
|
"hand_finger_ring_left_4",
|
|
"hand_finger_pinky_left_1",
|
|
"hand_finger_pinky_left_2",
|
|
"hand_finger_pinky_left_3",
|
|
"hand_finger_pinky_left_4",
|
|
"hand_wrist_right",
|
|
"hand_finger_thumb_right_1",
|
|
"hand_finger_thumb_right_2",
|
|
"hand_finger_thumb_right_3",
|
|
"hand_finger_thumb_right_4",
|
|
"hand_finger_index_right_1",
|
|
"hand_finger_index_right_2",
|
|
"hand_finger_index_right_3",
|
|
"hand_finger_index_right_4",
|
|
"hand_finger_middle_right_1",
|
|
"hand_finger_middle_right_2",
|
|
"hand_finger_middle_right_3",
|
|
"hand_finger_middle_right_4",
|
|
"hand_finger_ring_right_1",
|
|
"hand_finger_ring_right_2",
|
|
"hand_finger_ring_right_3",
|
|
"hand_finger_ring_right_4",
|
|
"hand_finger_pinky_right_1",
|
|
"hand_finger_pinky_right_2",
|
|
"hand_finger_pinky_right_3",
|
|
"hand_finger_pinky_right_4",
|
|
]
|
|
)
|
|
joint_names_2d.extend(
|
|
[
|
|
"hip_middle",
|
|
"shoulder_middle",
|
|
"head",
|
|
]
|
|
)
|
|
joint_names_3d = list(joint_names_2d)
|
|
|
|
main_limbs = [
|
|
("shoulder_left", "elbow_left"),
|
|
("elbow_left", "wrist_left"),
|
|
("shoulder_right", "elbow_right"),
|
|
("elbow_right", "wrist_right"),
|
|
("hip_left", "knee_left"),
|
|
("knee_left", "ankle_left"),
|
|
("hip_right", "knee_right"),
|
|
("knee_right", "ankle_right"),
|
|
]
|
|
|
|
# ==================================================================================================
|
|
|
|
|
|
def update_sample(sample, new_dir=""):
|
|
sample = copy.deepcopy(sample)
|
|
|
|
# Rename image paths
|
|
sample["imgpaths"] = [
|
|
os.path.join(new_dir, os.path.basename(v)) for v in sample["imgpaths"]
|
|
]
|
|
|
|
# Add placeholders for missing keys
|
|
sample["cameras_color"] = sample["cameras"]
|
|
sample["imgpaths_color"] = sample["imgpaths"]
|
|
sample["cameras_depth"] = []
|
|
|
|
return sample
|
|
|
|
|
|
# ==================================================================================================
|
|
|
|
|
|
def load_image(path: str):
|
|
image = cv2.imread(path, 3)
|
|
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
|
|
image = np.array(image, dtype=np.float32)
|
|
return image
|
|
|
|
|
|
# ==================================================================================================
|
|
|
|
|
|
def filter_poses(poses3D, poses2D, roomparams, joint_names, drop_few_limbs=True):
|
|
drop = []
|
|
for i, pose in enumerate(poses3D):
|
|
pose = np.array(pose)
|
|
valid_joints = [j for j in pose if j[-1] > 0.1]
|
|
|
|
# Drop persons with too few joints
|
|
if np.sum(pose[..., -1] > 0.1) < 5:
|
|
drop.append(i)
|
|
continue
|
|
|
|
# Drop too large or too small persons
|
|
mins = np.min(valid_joints, axis=0)
|
|
maxs = np.max(valid_joints, axis=0)
|
|
diff = maxs - mins
|
|
if any(((d > 2.3) for d in diff)):
|
|
drop.append(i)
|
|
continue
|
|
if all(((d < 0.3) for d in diff)):
|
|
drop.append(i)
|
|
continue
|
|
if (
|
|
(diff[0] < 0.2 and diff[1] < 0.2)
|
|
or (diff[1] < 0.2 and diff[2] < 0.2)
|
|
or (diff[2] < 0.2 and diff[0] < 0.2)
|
|
):
|
|
drop.append(i)
|
|
continue
|
|
|
|
# Drop persons outside room
|
|
mean = np.mean(valid_joints, axis=0)
|
|
mins = np.min(valid_joints, axis=0)
|
|
maxs = np.max(valid_joints, axis=0)
|
|
rsize = [r / 2 for r in roomparams["room_size"]]
|
|
rcent = roomparams["room_center"]
|
|
if any(
|
|
(
|
|
# Center of mass outside room
|
|
mean[j] > rsize[j] + rcent[j] or mean[j] < -rsize[j] + rcent[j]
|
|
for j in range(3)
|
|
)
|
|
) or any(
|
|
(
|
|
# One limb more than 10cm outside room
|
|
maxs[j] > rsize[j] + rcent[j] + 0.1
|
|
or mins[j] < -rsize[j] + rcent[j] - 0.1
|
|
for j in range(3)
|
|
)
|
|
):
|
|
drop.append(i)
|
|
continue
|
|
|
|
if drop_few_limbs:
|
|
# Drop persons with less than 3 limbs
|
|
found_limbs = 0
|
|
for limb in main_limbs:
|
|
start_idx = joint_names.index(limb[0])
|
|
end_idx = joint_names.index(limb[1])
|
|
if pose[start_idx, -1] > 0.1 and pose[end_idx, -1] > 0.1:
|
|
found_limbs += 1
|
|
if found_limbs < 3:
|
|
drop.append(i)
|
|
continue
|
|
|
|
# Drop persons with too small or high average limb length
|
|
total_length = 0
|
|
total_limbs = 0
|
|
for limb in main_limbs:
|
|
start_idx = joint_names.index(limb[0])
|
|
end_idx = joint_names.index(limb[1])
|
|
if pose[start_idx, -1] < 0.1 or pose[end_idx, -1] < 0.1:
|
|
continue
|
|
limb_length = np.linalg.norm(pose[end_idx, :3] - pose[start_idx, :3])
|
|
total_length += limb_length
|
|
total_limbs += 1
|
|
if total_limbs == 0:
|
|
drop.append(i)
|
|
continue
|
|
average_length = total_length / total_limbs
|
|
if average_length < 0.1:
|
|
drop.append(i)
|
|
continue
|
|
if average_length > 0.5:
|
|
drop.append(i)
|
|
continue
|
|
|
|
new_poses3D = []
|
|
new_poses2D = [[] for _ in range(len(poses2D))]
|
|
for i in range(len(poses3D)):
|
|
if len(poses3D[i]) != len(joint_names):
|
|
# Sometimes some joints of a poor detection are missing
|
|
continue
|
|
|
|
if i not in drop:
|
|
new_poses3D.append(poses3D[i])
|
|
for j in range(len(poses2D)):
|
|
new_poses2D[j].append(poses2D[j][i])
|
|
else:
|
|
new_pose = np.array(poses3D[i])
|
|
new_pose[..., -1] = 0.001
|
|
new_poses3D.append(new_pose)
|
|
for j in range(len(poses2D)):
|
|
new_pose = np.array(poses2D[j][i])
|
|
new_pose[..., -1] = 0.001
|
|
new_poses2D[j].append(new_pose)
|
|
|
|
new_poses3D = np.array(new_poses3D)
|
|
new_poses2D = np.array(new_poses2D)
|
|
if new_poses3D.size == 0:
|
|
new_poses3D = np.zeros([1, len(joint_names), 4])
|
|
new_poses2D = np.zeros([len(poses2D), 1, len(joint_names), 3])
|
|
|
|
return new_poses3D, new_poses2D
|
|
|
|
|
|
# ==================================================================================================
|
|
|
|
|
|
def update_keypoints(poses_2d: list, joint_names: List[str]) -> list:
|
|
new_views = []
|
|
for view in poses_2d:
|
|
new_bodies = []
|
|
for body in view:
|
|
body = body.tolist()
|
|
|
|
new_body = body[:17]
|
|
if whole_body["foots"]:
|
|
new_body.extend(body[17:22])
|
|
if whole_body["face"]:
|
|
new_body.extend(body[22:90])
|
|
if whole_body["hands"]:
|
|
new_body.extend(body[90:])
|
|
body = new_body
|
|
|
|
hlid = joint_names.index("hip_left")
|
|
hrid = joint_names.index("hip_right")
|
|
mid_hip = [
|
|
float(((body[hlid][0] + body[hrid][0]) / 2.0)),
|
|
float(((body[hlid][1] + body[hrid][1]) / 2.0)),
|
|
min(body[hlid][2], body[hrid][2]),
|
|
]
|
|
body.append(mid_hip)
|
|
|
|
slid = joint_names.index("shoulder_left")
|
|
srid = joint_names.index("shoulder_right")
|
|
mid_shoulder = [
|
|
float(((body[slid][0] + body[srid][0]) / 2.0)),
|
|
float(((body[slid][1] + body[srid][1]) / 2.0)),
|
|
min(body[slid][2], body[srid][2]),
|
|
]
|
|
body.append(mid_shoulder)
|
|
|
|
elid = joint_names.index("ear_left")
|
|
erid = joint_names.index("ear_right")
|
|
head = [
|
|
float(((body[elid][0] + body[erid][0]) / 2.0)),
|
|
float(((body[elid][1] + body[erid][1]) / 2.0)),
|
|
min(body[elid][2], body[erid][2]),
|
|
]
|
|
body.append(head)
|
|
|
|
new_bodies.append(body)
|
|
new_views.append(new_bodies)
|
|
|
|
return new_views
|
|
|
|
|
|
# ==================================================================================================
|
|
|
|
|
|
def main():
|
|
if any((whole_body[k] for k in whole_body)):
|
|
kpt_model = utils_2d_pose.load_wb_model()
|
|
else:
|
|
kpt_model = utils_2d_pose.load_model()
|
|
|
|
# Manually set matplotlib backend
|
|
matplotlib.use("TkAgg")
|
|
|
|
for dirname in sorted(os.listdir(test_img_dir)):
|
|
dirpath = os.path.join(test_img_dir, dirname)
|
|
|
|
if not os.path.isdir(dirpath):
|
|
continue
|
|
|
|
if (dirname[0] not in ["p", "h"]) or len(dirname) != 2:
|
|
continue
|
|
|
|
# Load sample infos
|
|
with open(os.path.join(dirpath, "sample.json"), "r", encoding="utf-8") as file:
|
|
sample = json.load(file)
|
|
sample = update_sample(sample, dirpath)
|
|
|
|
camparams = sample["cameras_color"]
|
|
roomparams = {
|
|
"room_size": sample["room_size"],
|
|
"room_center": sample["room_center"],
|
|
}
|
|
|
|
# Load color images
|
|
images_2d = []
|
|
for i in range(len(sample["cameras_color"])):
|
|
imgpath = sample["imgpaths_color"][i]
|
|
img = load_image(imgpath)
|
|
images_2d.append(img)
|
|
|
|
# Get 2D poses
|
|
poses_2d = utils_2d_pose.get_2d_pose(kpt_model, images_2d)
|
|
poses_2d = update_keypoints(poses_2d, joint_names_2d)
|
|
|
|
fig1 = draw_utils.show_poses2d(
|
|
poses_2d, np.array(images_2d), joint_names_2d, "2D detections"
|
|
)
|
|
fig1.savefig(os.path.join(dirpath, "2d-k.png"), dpi=fig1.dpi)
|
|
# draw_utils.utils_view.show_plots()
|
|
|
|
if len(images_2d) == 1:
|
|
draw_utils.utils_view.show_plots()
|
|
continue
|
|
|
|
# Get 3D poses
|
|
if sum(np.sum(p) for p in poses_2d) == 0:
|
|
poses3D = np.zeros([1, len(joint_names_3d), 4])
|
|
poses2D = np.zeros([len(images_2d), 1, len(joint_names_3d), 3])
|
|
else:
|
|
poses3D = triangulate_poses.get_3d_pose(
|
|
poses_2d, camparams, roomparams, joint_names_2d
|
|
)
|
|
poses2D = []
|
|
for cam in camparams:
|
|
poses_2d, _ = utils_pose.project_poses(poses3D, cam)
|
|
poses2D.append(poses_2d)
|
|
poses3D, poses2D = filter_poses(
|
|
poses3D,
|
|
poses2D,
|
|
roomparams,
|
|
joint_names_3d,
|
|
)
|
|
|
|
print("\n" + dirpath)
|
|
print(poses3D)
|
|
# print(poses2D)
|
|
|
|
fig2 = draw_utils.utils_view.show_poses3d(
|
|
poses3D, joint_names_3d, roomparams, camparams
|
|
)
|
|
fig3 = draw_utils.show_poses2d(
|
|
poses2D, np.array(images_2d), joint_names_3d, "2D reprojections"
|
|
)
|
|
fig2.savefig(os.path.join(dirpath, "3d-p.png"), dpi=fig2.dpi)
|
|
fig3.savefig(os.path.join(dirpath, "2d-p.png"), dpi=fig3.dpi)
|
|
draw_utils.utils_view.show_plots()
|
|
|
|
|
|
# ==================================================================================================
|
|
|
|
if __name__ == "__main__":
|
|
main()
|