SignVerse-2M / utils /draw_dw_lib.py
FangSen9000's picture
Update DWPose visualization styles (#2)
9e31a93
import math
import numpy as np
import matplotlib
import cv2
import sys
import os
import _pickle as cPickle
import gzip
import subprocess
import torch
import colorsys
from typing import List, Dict, Any, Optional, Tuple
eps = 0.01
RENDER_STYLES = {
"bold": {"line_scale": 3.0, "point_scale": 2.0},
"thin": {"line_scale": 1.0, "point_scale": 1.0},
}
def get_render_style(style: str = "bold") -> Dict[str, float]:
"""Return named DWPose rendering style parameters."""
if style not in RENDER_STYLES:
raise ValueError(f"Unknown render style '{style}'. Choose from: {sorted(RENDER_STYLES)}")
return RENDER_STYLES[style]
def filter_pose_by_confidence(
pose_data: Dict[str, Any],
conf_threshold: float = 0.6,
) -> Dict[str, Any]:
"""Filter low-confidence joints before rendering."""
filtered = {}
for key, value in pose_data.items():
if isinstance(value, np.ndarray):
filtered[key] = value.copy()
else:
filtered[key] = value
bodies = filtered.get("bodies")
body_scores = filtered.get("body_scores")
if bodies is not None:
bodies = np.array(bodies, copy=True)
min_valid = 1e-6
coord_mask = (bodies[:, 0] > min_valid) & (bodies[:, 1] > min_valid)
conf_mask = None
if body_scores is not None:
score_vec = np.asarray(body_scores).reshape(-1).astype(float)
conf_mask = score_vec < conf_threshold
if conf_mask.shape[0] < bodies.shape[0]:
conf_mask = np.pad(
conf_mask,
(0, bodies.shape[0] - conf_mask.shape[0]),
constant_values=False,
)
elif conf_mask.shape[0] > bodies.shape[0]:
conf_mask = conf_mask[: bodies.shape[0]]
valid_mask = coord_mask if conf_mask is None else coord_mask & (~conf_mask)
bodies[~valid_mask, :] = 0
filtered["bodies"] = bodies
if body_scores is not None:
subset = np.array(body_scores, copy=True)
if subset.ndim == 1:
subset = subset.reshape(1, -1)
else:
subset = np.arange(bodies.shape[0], dtype=float).reshape(1, -1)
if subset.shape[1] < bodies.shape[0]:
subset = np.pad(
subset,
((0, 0), (0, bodies.shape[0] - subset.shape[1])),
constant_values=-1,
)
elif subset.shape[1] > bodies.shape[0]:
subset = subset[:, : bodies.shape[0]]
subset[:, ~valid_mask] = -1
filtered["body_scores"] = subset
hands = filtered.get("hands")
hand_scores = filtered.get("hands_scores")
if hands is not None and hand_scores is not None:
hands = np.array(hands, copy=True)
scores = np.array(hand_scores)
if hands.ndim == 3 and scores.ndim in (2, 3):
for h in range(hands.shape[0]):
cur_scores = scores[h] if scores.ndim == 2 else scores[h]
mask = (cur_scores < conf_threshold) | (cur_scores <= 0)
hands[h][mask, :] = 0
elif hands.ndim == 2 and scores.ndim in (1, 2):
cur_scores = scores if scores.ndim == 1 else scores.reshape(-1)
mask = (cur_scores < conf_threshold) | (cur_scores <= 0)
hands[mask, :] = 0
filtered["hands"] = hands
faces = filtered.get("faces")
face_scores = filtered.get("faces_scores")
if faces is not None and face_scores is not None:
faces = np.array(faces, copy=True)
scores = np.array(face_scores)
if faces.ndim == 3 and scores.ndim in (2, 3):
for f in range(faces.shape[0]):
cur_scores = scores[f] if scores.ndim == 2 else scores[f]
mask = (cur_scores < conf_threshold) | (cur_scores <= 0)
faces[f][mask, :] = 0
elif faces.ndim == 2 and scores.ndim in (1, 2):
cur_scores = scores if scores.ndim == 1 else scores.reshape(-1)
mask = (cur_scores < conf_threshold) | (cur_scores <= 0)
faces[mask, :] = 0
filtered["faces"] = faces
return filtered
def alpha_blend_color(color, alpha):
"""blend color according to point conf
"""
return [int(c * alpha) for c in color]
def draw_bodypose(canvas, candidate, subset, score, transparent=False, line_scale=1.0, point_scale=1.0):
"""Draw body pose on canvas
Args:
canvas: numpy array canvas to draw on
candidate: pose candidate
subset: pose subset
score: confidence scores
transparent: whether to use transparent background
Returns:
canvas: drawn canvas
"""
H, W, C = canvas.shape
candidate = np.array(candidate)
subset = np.array(subset)
stickwidth = max(1, int(round(4 * line_scale)))
limbSeq = [[2, 3], [2, 6], [3, 4], [4, 5], [6, 7], [7, 8], [2, 9], [9, 10],
[10, 11], [2, 12], [12, 13], [13, 14], [2, 1], [1, 15], [15, 17],
[1, 16], [16, 18], [3, 17], [6, 18]]
colors = [[255, 0, 0], [255, 85, 0], [255, 170, 0], [255, 255, 0], [170, 255, 0], [85, 255, 0], [0, 255, 0],
[0, 255, 85], [0, 255, 170], [0, 255, 255], [0, 170, 255], [0, 85, 255], [0, 0, 255], [85, 0, 255],
[170, 0, 255], [255, 0, 255], [255, 0, 170], [255, 0, 85]]
# Add alpha channel if transparent
if transparent:
colors = [color + [255] for color in colors]
for i in range(17):
for n in range(len(subset)):
index = subset[n][np.array(limbSeq[i]) - 1]
conf = score[n][np.array(limbSeq[i]) - 1]
if conf[0] < 0.3 or conf[1] < 0.3:
continue
Y = candidate[index.astype(int), 0] * float(W)
X = candidate[index.astype(int), 1] * float(H)
mX = np.mean(X)
mY = np.mean(Y)
length = ((X[0] - X[1]) ** 2 + (Y[0] - Y[1]) ** 2) ** 0.5
angle = math.degrees(math.atan2(X[0] - X[1], Y[0] - Y[1]))
polygon = cv2.ellipse2Poly((int(mY), int(mX)), (int(length / 2), stickwidth), int(angle), 0, 360, 1)
if transparent:
color = colors[i][:-1] + [int(255 * conf[0] * conf[1])] # Adjust alpha based on confidence
else:
color = colors[i]
cv2.fillConvexPoly(canvas, polygon, color)
canvas = (canvas * 0.6).astype(np.uint8)
for i in range(18):
for n in range(len(subset)):
index = int(subset[n][i])
if index == -1:
continue
x, y = candidate[index][0:2]
conf = score[n][i]
x = int(x * W)
y = int(y * H)
if transparent:
color = colors[i][:-1] + [int(255 * conf)] # Adjust alpha based on confidence
else:
color = colors[i]
cv2.circle(canvas, (int(x), int(y)), max(1, int(round(4 * point_scale))), color, thickness=-1)
return canvas
def draw_handpose(canvas, all_hand_peaks, all_hand_scores, transparent=False, line_scale=1.0, point_scale=1.0):
"""Draw hand pose on canvas"""
H, W, C = canvas.shape
edges = [[0, 1], [1, 2], [2, 3], [3, 4], [0, 5], [5, 6], [6, 7], [7, 8], [0, 9], [9, 10],
[10, 11], [11, 12], [0, 13], [13, 14], [14, 15], [15, 16], [0, 17], [17, 18], [18, 19], [19, 20]]
for peaks, scores in zip(all_hand_peaks, all_hand_scores):
for ie, e in enumerate(edges):
x1, y1 = peaks[e[0]]
x2, y2 = peaks[e[1]]
x1 = int(x1 * W)
y1 = int(y1 * H)
x2 = int(x2 * W)
y2 = int(y2 * H)
score = scores[e[0]] * scores[e[1]]
if x1 > eps and y1 > eps and x2 > eps and y2 > eps:
color = matplotlib.colors.hsv_to_rgb([ie / float(len(edges)), 1.0, 1.0])
if transparent:
color = tuple(int(c * 255) for c in color) + (int(score * 255),)
else:
color = tuple(int(c * score * 255) for c in color)
cv2.line(canvas, (x1, y1), (x2, y2), color, thickness=max(1, int(round(2 * line_scale))))
for i, keypoint in enumerate(peaks):
x, y = keypoint
x = int(x * W)
y = int(y * H)
if x > eps and y > eps:
if transparent:
color = (0, 0, 0, int(scores[i] * 255))
else:
color = (0, 0, int(scores[i] * 255))
cv2.circle(canvas, (x, y), max(1, int(round(4 * point_scale))), color, thickness=-1)
return canvas
def draw_facepose(canvas, all_lmks, all_scores, transparent=False, point_scale=1.0):
"""Draw face pose on canvas"""
H, W, C = canvas.shape
for lmks, scores in zip(all_lmks, all_scores):
for lmk, score in zip(lmks, scores):
x, y = lmk
x = int(x * W)
y = int(y * H)
if x > eps and y > eps:
if transparent:
color = (255, 255, 255, int(score * 255)) # White with alpha
else:
conf = int(score * 255)
color = (conf, conf, conf) # Original grayscale
cv2.circle(canvas, (x, y), max(1, int(round(3 * point_scale))), color, thickness=-1)
return canvas
def draw_pose(
pose,
H,
W,
include_body=True,
include_hand=True,
include_face=True,
ref_w=2160,
transparent=False,
line_scale=1.0,
point_scale=1.0,
):
"""vis dwpose outputs with optional transparent background
Args:
pose (Dict): DWposeDetector outputs - 支持新的person_id格式和旧格式
H (int): height
W (int): width
include_body (bool): whether to draw body keypoints
include_hand (bool): whether to draw hand keypoints
include_face (bool): whether to draw face keypoints
ref_w (int, optional): reference width. Defaults to 2160.
transparent (bool, optional): whether to use transparent background. Defaults to False.
Returns:
np.ndarray: image pixel value in RGBA mode if transparent=True, otherwise RGB mode
"""
sz = min(H, W)
sr = (ref_w / sz) if sz != ref_w else 1
# Create canvas - now with alpha channel if transparent
if transparent:
canvas = np.zeros(shape=(int(H*sr), int(W*sr), 4), dtype=np.uint8)
else:
canvas = np.zeros(shape=(int(H*sr), int(W*sr), 3), dtype=np.uint8)
# 检查是否是新的person_id数据格式
if 'num_persons' in pose and pose['num_persons'] > 0:
# 使用新的多人数据结构
processed_data = process_pose_data(pose, H, W)
bodies = processed_data['bodies']
faces = processed_data['faces']
hands = processed_data['hands']
candidate = bodies['candidate']
subset = bodies['subset']
if include_body:
canvas = draw_bodypose(
canvas,
candidate,
subset,
score=bodies['score'],
transparent=transparent,
line_scale=line_scale,
point_scale=point_scale,
)
if include_hand:
canvas = draw_handpose(
canvas,
hands,
processed_data['hands_score'],
transparent=transparent,
line_scale=line_scale,
point_scale=point_scale,
)
if include_face:
canvas = draw_facepose(
canvas,
faces,
processed_data['faces_score'],
transparent=transparent,
point_scale=point_scale,
)
else:
# 兼容旧的数据格式 - 作为备选方案
try:
processed_data = process_pose_data(pose, H, W)
bodies = processed_data['bodies']
faces = processed_data['faces']
hands = processed_data['hands']
candidate = bodies['candidate']
subset = bodies['subset']
if include_body:
canvas = draw_bodypose(
canvas,
candidate,
subset,
score=bodies['score'],
transparent=transparent,
line_scale=line_scale,
point_scale=point_scale,
)
if include_hand:
canvas = draw_handpose(
canvas,
hands,
processed_data['hands_score'],
transparent=transparent,
line_scale=line_scale,
point_scale=point_scale,
)
if include_face:
canvas = draw_facepose(
canvas,
faces,
processed_data['faces_score'],
transparent=transparent,
point_scale=point_scale,
)
except Exception as e:
print(f"绘制旧格式数据失败: {str(e)}")
# 返回空画布
pass
if transparent:
return cv2.cvtColor(cv2.resize(canvas, (W, H)), cv2.COLOR_BGRA2RGBA).transpose(2, 0, 1)
else:
return cv2.cvtColor(cv2.resize(canvas, (W, H)), cv2.COLOR_BGR2RGB).transpose(2, 0, 1)
def process_pose_data(pose_data: Dict[str, Any], height: int, width: int) -> Dict[str, Any]:
"""
处理姿势数据,完全支持新的person_id数据结构
"""
processed_data = {}
# 确保使用新的数据结构
if 'num_persons' in pose_data and pose_data['num_persons'] > 0:
num_persons = pose_data['num_persons']
# 收集所有人的关键点数据
all_bodies = []
all_body_scores = []
all_hands = []
all_hand_scores = []
all_faces = []
all_face_scores = []
for person_id in range(num_persons):
person_key = f'person_{person_id}'
if person_key in pose_data:
person_data = pose_data[person_key]
all_bodies.append(person_data['body_keypoints'])
all_body_scores.append(person_data['body_scores'])
all_hands.extend([person_data['left_hand_keypoints'], person_data['right_hand_keypoints']])
all_hand_scores.extend([person_data['left_hand_scores'], person_data['right_hand_scores']])
all_faces.append(person_data['face_keypoints'])
all_face_scores.append(person_data['face_scores'])
# 合并所有人的数据
if all_bodies:
bodies = np.vstack(all_bodies)
body_scores = np.array(all_body_scores)
# 创建subset - 为每个人创建独立的subset行
subset = []
for person_id in range(num_persons):
person_subset = list(range(person_id * 18, (person_id + 1) * 18))
subset.append(person_subset)
subset = np.array(subset)
# 创建scores - 基于body_scores中的有效性
scores = np.ones_like(body_scores)
for i in range(num_persons):
for j in range(18):
if body_scores[i, j] < 0: # 如果body_scores为负数,认为无效
scores[i, j] = 0.0
else:
scores[i, j] = 1.0
else:
bodies = np.array([])
subset = np.array([[]])
scores = np.array([[]])
hands = np.array(all_hands) if all_hands else np.array([])
hand_scores = np.array(all_hand_scores) if all_hand_scores else np.array([])
faces = np.array(all_faces) if all_faces else np.array([])
face_scores = np.array(all_face_scores) if all_face_scores else np.array([])
else:
# 兼容旧的单人数据格式
raw_bodies = np.array(pose_data.get('bodies', []), dtype=np.float32)
raw_body_scores = np.array(pose_data.get('body_scores', []), dtype=np.float32)
raw_hands = np.array(pose_data.get('hands', []), dtype=np.float32)
raw_hand_scores = np.array(pose_data.get('hands_scores', []), dtype=np.float32)
raw_faces = np.array(pose_data.get('faces', []), dtype=np.float32)
raw_face_scores = np.array(pose_data.get('faces_scores', []), dtype=np.float32)
if raw_bodies.size > 0:
bodies = raw_bodies.reshape(-1, 2)
subset = np.arange(bodies.shape[0], dtype=np.int32)[None, :]
if raw_body_scores.size > 0:
scores = raw_body_scores.reshape(1, -1)
else:
scores = np.ones((1, bodies.shape[0]), dtype=np.float32)
else:
bodies = np.array([])
subset = np.array([[]])
scores = np.array([[]])
hands = raw_hands if raw_hands.size > 0 else np.array([])
hand_scores = raw_hand_scores if raw_hand_scores.size > 0 else np.array([])
faces = raw_faces if raw_faces.size > 0 else np.array([])
face_scores = raw_face_scores if raw_face_scores.size > 0 else np.array([])
processed_data['bodies'] = {
'candidate': bodies,
'subset': subset,
'score': scores
}
processed_data['hands'] = hands
processed_data['hands_score'] = hand_scores
processed_data['faces'] = faces
processed_data['faces_score'] = face_scores
return processed_data