Spaces:
Running on Zero
Running on Zero
File size: 11,160 Bytes
49d36c0 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 | # SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
"""Shared 2D keypoint visualisation utilities."""
import functools
import math
from pathlib import Path
import cv2
import numpy as np
import torch
from tqdm import tqdm
from gem.utils.pylogger import Log
from gem.utils.video_io_utils import get_video_lwh, get_video_reader
# PyTorch >=2.6 defaults weights_only=True, but our saved files contain numpy arrays.
_torch_load = functools.partial(torch.load, weights_only=False)
# ---------------------------------------------------------------------------
# Constants
# ---------------------------------------------------------------------------
PARENTS_77 = [
-1,
0,
1,
2,
3,
4,
5,
6,
6,
6,
6,
3,
11,
12,
13,
14,
15,
16,
17,
14,
19,
20,
21,
22,
14,
24,
25,
26,
27,
14,
29,
30,
31,
32,
14,
34,
35,
36,
37,
3,
39,
40,
41,
42,
43,
44,
45,
42,
47,
48,
49,
50,
42,
52,
53,
54,
55,
42,
57,
58,
59,
60,
42,
62,
63,
64,
65,
0,
67,
68,
69,
70,
0,
72,
73,
74,
75,
]
COCO_SKELETON = [
[15, 13],
[13, 11],
[16, 14],
[14, 12],
[11, 12],
[5, 11],
[6, 12],
[5, 6],
[5, 7],
[6, 8],
[7, 9],
[8, 10],
[1, 2],
[0, 1],
[0, 2],
[1, 3],
[2, 4],
[3, 5],
[4, 6],
]
# ---------------------------------------------------------------------------
# color palette (BGR for OpenCV)
# ---------------------------------------------------------------------------
_PART_COLORS_77 = {
"torso": (0, 215, 255), # gold
"head": (180, 130, 255), # light purple
"left_arm": (0, 255, 100), # green
"left_hand": (130, 255, 130), # light green
"right_arm": (50, 130, 255), # orange
"right_hand": (130, 180, 255), # light orange/salmon
"left_leg": (255, 190, 0), # cyan-blue
"right_leg": (255, 0, 170), # magenta
}
# Map each of the 77 joints to its body-part group
_JOINT_GROUP_77 = [""] * 77
for _j in range(0, 4):
_JOINT_GROUP_77[_j] = "torso"
for _j in range(4, 11):
_JOINT_GROUP_77[_j] = "head"
for _j in range(11, 14):
_JOINT_GROUP_77[_j] = "left_arm"
for _j in range(14, 39):
_JOINT_GROUP_77[_j] = "left_hand"
for _j in range(39, 42):
_JOINT_GROUP_77[_j] = "right_arm"
for _j in range(42, 67):
_JOINT_GROUP_77[_j] = "right_hand"
for _j in range(67, 72):
_JOINT_GROUP_77[_j] = "left_leg"
for _j in range(72, 77):
_JOINT_GROUP_77[_j] = "right_leg"
# Bone stickwidth (ellipse minor-axis): body/limb = 4, hands = 2, rest = 3
_BONE_STICKWIDTH_77 = [3] * 77
for _j in range(0, 4):
_BONE_STICKWIDTH_77[_j] = 4 # torso
for _j in range(67, 77):
_BONE_STICKWIDTH_77[_j] = 4 # legs
for _j in range(14, 39):
_BONE_STICKWIDTH_77[_j] = 2 # left hand
for _j in range(42, 67):
_BONE_STICKWIDTH_77[_j] = 2 # right hand
# Joint radius: body/limbs = 4, hands = 2
_JOINT_RADIUS_77 = [4] * 77
for _j in range(14, 39):
_JOINT_RADIUS_77[_j] = 2
for _j in range(42, 67):
_JOINT_RADIUS_77[_j] = 2
# Per-bone colors for the 19 COCO skeleton bones (BGR)
_COCO_BONE_COLORS = [
(255, 190, 0), # 0: 15-13 left leg
(255, 190, 0), # 1: 13-11 left leg
(255, 0, 170), # 2: 16-14 right leg
(255, 0, 170), # 3: 14-12 right leg
(0, 215, 255), # 4: 11-12 hip
(0, 215, 255), # 5: 5-11 left torso
(0, 215, 255), # 6: 6-12 right torso
(0, 215, 255), # 7: 5-6 shoulders
(0, 255, 100), # 8: 5-7 left upper arm
(50, 130, 255), # 9: 6-8 right upper arm
(0, 255, 100), # 10: 7-9 left forearm
(50, 130, 255), # 11: 8-10 right forearm
(180, 130, 255), # 12: 1-2 eyes
(180, 130, 255), # 13: 0-1 nose-left eye
(180, 130, 255), # 14: 0-2 nose-right eye
(180, 130, 255), # 15: 1-3 left ear
(180, 130, 255), # 16: 2-4 right ear
(0, 255, 100), # 17: 3-5 left ear-shoulder
(50, 130, 255), # 18: 4-6 right ear-shoulder
]
# Per-joint colors for 17 COCO joints
_COCO_JOINT_COLORS = [
(180, 130, 255), # 0: nose
(180, 130, 255), # 1: left eye
(180, 130, 255), # 2: right eye
(180, 130, 255), # 3: left ear
(180, 130, 255), # 4: right ear
(0, 255, 100), # 5: left shoulder
(50, 130, 255), # 6: right shoulder
(0, 255, 100), # 7: left elbow
(50, 130, 255), # 8: right elbow
(0, 255, 100), # 9: left wrist
(50, 130, 255), # 10: right wrist
(255, 190, 0), # 11: left hip
(255, 0, 170), # 12: right hip
(255, 190, 0), # 13: left knee
(255, 0, 170), # 14: right knee
(255, 190, 0), # 15: left ankle
(255, 0, 170), # 16: right ankle
]
# ---------------------------------------------------------------------------
# Helpers
# ---------------------------------------------------------------------------
def smooth_bbx_xyxy(bbx_xyxy, window=5):
"""Apply moving-average smoothing to a (L, 4) bounding-box sequence."""
if bbx_xyxy.shape[0] <= window:
return bbx_xyxy
kernel = torch.ones(1, 1, window, dtype=bbx_xyxy.dtype) / window
# (L, 4) -> (4, 1, L) for conv1d, then back
padded = bbx_xyxy.T.unsqueeze(1) # (4, 1, L)
pad_size = window // 2
padded = torch.nn.functional.pad(padded, (pad_size, pad_size), mode="replicate")
smoothed = torch.nn.functional.conv1d(padded, kernel).squeeze(1).T # (L, 4)
return smoothed
def _open_cv2_writer(path, width, height, fps):
"""Open an OpenCV VideoWriter for mp4v output."""
Path(path).parent.mkdir(parents=True, exist_ok=True)
fourcc = cv2.VideoWriter_fourcc(*"mp4v")
return cv2.VideoWriter(str(path), fourcc, float(fps), (int(width), int(height)))
def _draw_ellipse_bone(canvas, pt1, pt2, color, stickwidth):
"""Draw a bone as a filled ellipse between two joints."""
x1, y1 = pt1
x2, y2 = pt2
mx, my = (x1 + x2) / 2, (y1 + y2) / 2
length = math.hypot(x1 - x2, y1 - y2)
if length < 1:
return
angle = math.degrees(math.atan2(y1 - y2, x1 - x2))
polygon = cv2.ellipse2Poly(
(int(mx), int(my)), (int(length / 2), stickwidth), int(angle), 0, 360, 1
)
cv2.fillConvexPoly(canvas, polygon, color, lineType=cv2.LINE_AA)
# ---------------------------------------------------------------------------
# Main visualisation
# ---------------------------------------------------------------------------
def render_2d_keypoints(video_path, vitpose_path, bbx_path, output_path, fps=30, conf_thr=0.5):
"""Render 2D keypoint overlay on a video.
Parameters
----------
video_path : str or Path
Path to the input video file.
vitpose_path : str or Path
Path to the vitpose ``.pt`` file (tensor of shape ``(L, J, 2/3)``).
bbx_path : str or Path
Path to the bounding-box ``.pt`` file.
output_path : str or Path
Destination path for the rendered overlay video.
fps : int
Frames per second for the output video.
conf_thr : float
Confidence threshold below which joints/bones are hidden.
"""
if not Path(vitpose_path).exists():
Log.info("[2D KP] Missing vitpose results. Skipping 2D keypoint render.")
return
vitpose = _torch_load(vitpose_path)
if isinstance(vitpose, tuple):
vitpose = vitpose[0]
if isinstance(vitpose, np.ndarray):
vitpose = torch.from_numpy(vitpose)
assert vitpose.ndim == 3 and vitpose.shape[-1] >= 2, "vitpose expected (L, J, 2/3)"
bbx = _torch_load(bbx_path)
bbx.get("detected_bbx_xys", bbx.get("bbx_xys", None))
reader = get_video_reader(video_path)
writer = _open_cv2_writer(
output_path, get_video_lwh(video_path)[1], get_video_lwh(video_path)[2], fps
)
for i, img_raw in tqdm(
enumerate(reader), total=get_video_lwh(video_path)[0], desc="Render 2D KP"
):
if i >= vitpose.shape[0]:
break
img = img_raw.copy()
keypoints = vitpose[i].cpu().numpy()
use_conf = keypoints.shape[1] == 3
num_joints = keypoints.shape[0]
if num_joints == 77:
# Draw bones as filled ellipses
for child_idx, parent_idx in enumerate(PARENTS_77):
if parent_idx < 0:
continue
if use_conf and (
keypoints[parent_idx][2] <= conf_thr or keypoints[child_idx][2] <= conf_thr
):
continue
pt1 = keypoints[parent_idx][:2].tolist()
pt2 = keypoints[child_idx][:2].tolist()
group = _JOINT_GROUP_77[child_idx]
color = _PART_COLORS_77[group]
stickwidth = _BONE_STICKWIDTH_77[child_idx]
cur_canvas = img.copy()
_draw_ellipse_bone(cur_canvas, pt1, pt2, color, stickwidth)
img = cv2.addWeighted(img, 0.4, cur_canvas, 0.6, 0)
# Draw joints with dark outline + colored fill
for j in range(num_joints):
if use_conf and keypoints[j][2] <= conf_thr:
continue
x, y = keypoints[j][:2].astype(int)
group = _JOINT_GROUP_77[j]
color = _PART_COLORS_77[group]
radius = _JOINT_RADIUS_77[j]
cv2.circle(img, (x, y), radius, (0, 0, 0), -1, cv2.LINE_AA)
cv2.circle(img, (x, y), max(radius - 1, 1), color, -1, cv2.LINE_AA)
elif num_joints == 17:
# Draw bones as filled ellipses
for bone_idx, (a, b) in enumerate(COCO_SKELETON):
if use_conf and (keypoints[a][2] <= conf_thr or keypoints[b][2] <= conf_thr):
continue
pt1 = keypoints[a][:2].tolist()
pt2 = keypoints[b][:2].tolist()
cur_canvas = img.copy()
_draw_ellipse_bone(cur_canvas, pt1, pt2, _COCO_BONE_COLORS[bone_idx], 4)
img = cv2.addWeighted(img, 0.4, cur_canvas, 0.6, 0)
# Draw joints with dark outline + colored fill
for j in range(num_joints):
if use_conf and keypoints[j][2] <= conf_thr:
continue
x, y = keypoints[j][:2].astype(int)
cv2.circle(img, (x, y), 4, (0, 0, 0), -1, cv2.LINE_AA)
cv2.circle(img, (x, y), 3, _COCO_JOINT_COLORS[j], -1, cv2.LINE_AA)
# # Draw bounding box
# if bbx_xys is not None and i < len(bbx_xys):
# cx, cy, s = bbx_xys[i].detach().cpu().numpy().tolist()
# half = 0.5 * float(s)
# bx0, by0 = int(round(cx - half)), int(round(cy - half))
# bx1, by1 = int(round(cx + half)), int(round(cy + half))
# cv2.rectangle(img, (bx0, by0), (bx1, by1), (0, 255, 255), 2, cv2.LINE_AA)
writer.write(img[..., ::-1])
writer.release()
reader.close()
Log.info(f"[2D KP] Saved overlay to {output_path}")
|