megalado
commited on
Commit
·
891dfcb
1
Parent(s):
5ee9026
Create different animations based on input parameters
Browse files
app.py
CHANGED
|
@@ -9,33 +9,7 @@ import subprocess
|
|
| 9 |
import glob
|
| 10 |
import requests
|
| 11 |
import time
|
| 12 |
-
|
| 13 |
-
def download_checkpoint():
|
| 14 |
-
"""Download the recommended checkpoint if not present"""
|
| 15 |
-
# Create checkpoints directory
|
| 16 |
-
os.makedirs("checkpoints", exist_ok=True)
|
| 17 |
-
|
| 18 |
-
# Define the target checkpoint path
|
| 19 |
-
checkpoint_path = "checkpoints/humanml_trans_enc_512.pt"
|
| 20 |
-
|
| 21 |
-
if not Path(checkpoint_path).exists():
|
| 22 |
-
print(f"Downloading checkpoint to {checkpoint_path}...")
|
| 23 |
-
# URL for the checkpoint from HuggingFace or direct link
|
| 24 |
-
url = "https://huggingface.co/spaces/mohaed/testMDM/resolve/main/checkpoints/mld_humanml.pt"
|
| 25 |
-
|
| 26 |
-
# Download the file
|
| 27 |
-
response = requests.get(url, stream=True)
|
| 28 |
-
response.raise_for_status()
|
| 29 |
-
|
| 30 |
-
with open(checkpoint_path, 'wb') as f:
|
| 31 |
-
for chunk in response.iter_content(chunk_size=8192):
|
| 32 |
-
f.write(chunk)
|
| 33 |
-
|
| 34 |
-
print(f"Checkpoint downloaded to {checkpoint_path}")
|
| 35 |
-
else:
|
| 36 |
-
print(f"Checkpoint already exists at {checkpoint_path}")
|
| 37 |
-
|
| 38 |
-
return checkpoint_path
|
| 39 |
|
| 40 |
def inspect_mdm_repo():
|
| 41 |
"""Check the structure of the MDM repository"""
|
|
@@ -45,54 +19,109 @@ def inspect_mdm_repo():
|
|
| 45 |
|
| 46 |
print("Inspecting MDM repository structure:")
|
| 47 |
# List top-level directories and files
|
| 48 |
-
for item in sorted(os.listdir("motion-diffusion-model")):
|
| 49 |
path = os.path.join("motion-diffusion-model", item)
|
| 50 |
if os.path.isdir(path):
|
| 51 |
print(f"Directory: {item}")
|
| 52 |
-
# List files in the directory
|
| 53 |
-
for subitem in sorted(os.listdir(path)):
|
| 54 |
print(f" - {subitem}")
|
| 55 |
else:
|
| 56 |
print(f"File: {item}")
|
|
|
|
| 57 |
|
| 58 |
-
def
|
| 59 |
-
"""Create a
|
| 60 |
-
print("Creating
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 61 |
|
| 62 |
# Create a simple visualization script
|
| 63 |
-
with open("
|
| 64 |
-
f.write("""
|
| 65 |
import numpy as np
|
| 66 |
import matplotlib.pyplot as plt
|
| 67 |
from matplotlib.animation import FuncAnimation
|
| 68 |
import os
|
| 69 |
from mpl_toolkits.mplot3d import Axes3D
|
|
|
|
| 70 |
|
| 71 |
-
#
|
| 72 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 73 |
joints = 16 # Number of joints in a simplified skeleton
|
| 74 |
dims = 3 # x, y, z
|
| 75 |
|
| 76 |
motion = np.zeros((frames, joints, dims))
|
| 77 |
|
| 78 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 79 |
for frame in range(frames):
|
| 80 |
t = frame / frames
|
| 81 |
|
| 82 |
# Basic forward motion
|
| 83 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 84 |
|
| 85 |
# Root joint (pelvis)
|
| 86 |
-
|
| 87 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 88 |
|
| 89 |
# Spine and head (joints 1, 2, 3)
|
| 90 |
for i in range(1, 4):
|
| 91 |
-
motion[frame, i, 1] = 0
|
| 92 |
motion[frame, i, 2] = motion[frame, 0, 2] + i * 0.2 # Stack joints vertically
|
| 93 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 94 |
# Left leg (joints 4, 5, 6)
|
| 95 |
-
|
|
|
|
| 96 |
motion[frame, 4, 1] = 0.2 # Left hip position
|
| 97 |
motion[frame, 4, 2] = motion[frame, 0, 2] - 0.1
|
| 98 |
motion[frame, 5, 1] = 0.2
|
|
@@ -101,7 +130,7 @@ def generate_simple_motion(frames=90):
|
|
| 101 |
motion[frame, 6, 2] = motion[frame, 5, 2] - 0.5 + swing_leg_l * 0.3
|
| 102 |
|
| 103 |
# Right leg (joints 7, 8, 9)
|
| 104 |
-
swing_leg_r = np.sin(t * 2 * np.pi *
|
| 105 |
motion[frame, 7, 1] = -0.2 # Right hip position
|
| 106 |
motion[frame, 7, 2] = motion[frame, 0, 2] - 0.1
|
| 107 |
motion[frame, 8, 1] = -0.2
|
|
@@ -110,17 +139,38 @@ def generate_simple_motion(frames=90):
|
|
| 110 |
motion[frame, 9, 2] = motion[frame, 8, 2] - 0.5 + swing_leg_r * 0.3
|
| 111 |
|
| 112 |
# Left arm (joints 10, 11, 12)
|
| 113 |
-
|
| 114 |
-
|
| 115 |
-
|
| 116 |
-
|
| 117 |
-
|
| 118 |
-
|
| 119 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 120 |
|
| 121 |
# Right arm (joints 13, 14, 15)
|
| 122 |
-
swing_arm_r = np.sin(t * 2 * np.pi *
|
| 123 |
-
motion[frame, 13, 1] = -0.3
|
| 124 |
motion[frame, 13, 2] = motion[frame, 3, 2] - 0.2
|
| 125 |
motion[frame, 14, 1] = -0.3 + swing_arm_r * 0.2
|
| 126 |
motion[frame, 14, 2] = motion[frame, 13, 2] - 0.4
|
|
@@ -131,7 +181,7 @@ def generate_simple_motion(frames=90):
|
|
| 131 |
|
| 132 |
def visualize_motion(output_path):
|
| 133 |
# Generate motion data
|
| 134 |
-
motion_data =
|
| 135 |
|
| 136 |
# Get dimensions
|
| 137 |
frames, joints, dims = motion_data.shape
|
|
@@ -153,9 +203,10 @@ def visualize_motion(output_path):
|
|
| 153 |
ax.clear()
|
| 154 |
|
| 155 |
# Set axis limits
|
| 156 |
-
|
| 157 |
-
ax.
|
| 158 |
-
ax.
|
|
|
|
| 159 |
|
| 160 |
# Set labels
|
| 161 |
ax.set_xlabel('X (forward)')
|
|
@@ -173,40 +224,40 @@ def visualize_motion(output_path):
|
|
| 173 |
[motion_data[frame, start, 1], motion_data[frame, end, 1]],
|
| 174 |
[motion_data[frame, start, 2], motion_data[frame, end, 2]], 'r-')
|
| 175 |
|
| 176 |
-
ax.set_title(f"
|
| 177 |
return ax
|
| 178 |
|
| 179 |
# Create animation
|
| 180 |
-
anim = FuncAnimation(fig, update, frames=motion_data.shape[0], interval=1000/30)
|
| 181 |
|
| 182 |
# Save animation
|
| 183 |
os.makedirs(os.path.dirname(output_path) or '.', exist_ok=True)
|
| 184 |
anim.save(output_path, writer='ffmpeg', fps=30)
|
| 185 |
plt.close()
|
| 186 |
|
| 187 |
-
print(f"Animation saved to {output_path}")
|
| 188 |
return output_path
|
| 189 |
|
| 190 |
-
# Create and visualize a
|
| 191 |
-
visualize_motion("
|
| 192 |
""")
|
| 193 |
|
| 194 |
# Run the script
|
| 195 |
-
subprocess.run(["python", "
|
| 196 |
|
| 197 |
-
if os.path.exists(
|
| 198 |
-
return
|
| 199 |
else:
|
| 200 |
return None
|
| 201 |
|
| 202 |
def text_to_motion(text_prompt, motion_length=3.0, seed=0):
|
| 203 |
"""Generate motion from text prompt using MDM"""
|
| 204 |
try:
|
| 205 |
-
# Inspect the MDM repository structure
|
| 206 |
inspect_mdm_repo()
|
| 207 |
|
| 208 |
-
#
|
| 209 |
-
return
|
| 210 |
|
| 211 |
except Exception as e:
|
| 212 |
print(f"Error generating motion: {str(e)}")
|
|
@@ -223,7 +274,7 @@ demo = gr.Interface(
|
|
| 223 |
],
|
| 224 |
outputs=gr.Video(label="Generated Motion"),
|
| 225 |
title="Motion Diffusion Model Demo",
|
| 226 |
-
description="Generate human motions from text descriptions"
|
| 227 |
)
|
| 228 |
|
| 229 |
# Launch the app
|
|
|
|
| 9 |
import glob
|
| 10 |
import requests
|
| 11 |
import time
|
| 12 |
+
import random
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
|
| 14 |
def inspect_mdm_repo():
|
| 15 |
"""Check the structure of the MDM repository"""
|
|
|
|
| 19 |
|
| 20 |
print("Inspecting MDM repository structure:")
|
| 21 |
# List top-level directories and files
|
| 22 |
+
for item in sorted(os.listdir("motion-diffusion-model"))[:5]: # Limit to first 5 for brevity
|
| 23 |
path = os.path.join("motion-diffusion-model", item)
|
| 24 |
if os.path.isdir(path):
|
| 25 |
print(f"Directory: {item}")
|
| 26 |
+
# List files in the directory (up to 3)
|
| 27 |
+
for subitem in sorted(os.listdir(path))[:3]:
|
| 28 |
print(f" - {subitem}")
|
| 29 |
else:
|
| 30 |
print(f"File: {item}")
|
| 31 |
+
print("...") # Indicate truncated output
|
| 32 |
|
| 33 |
+
def create_motion_animation(text_prompt, motion_length=3.0, seed=0):
|
| 34 |
+
"""Create a motion animation based on input parameters"""
|
| 35 |
+
print(f"Creating animation for: '{text_prompt}', length: {motion_length}s, seed: {seed}")
|
| 36 |
+
|
| 37 |
+
# Use the seed for reproducibility
|
| 38 |
+
np.random.seed(int(seed))
|
| 39 |
+
random.seed(int(seed))
|
| 40 |
+
|
| 41 |
+
# Parse the text prompt to influence the animation
|
| 42 |
+
walking = "walk" in text_prompt.lower()
|
| 43 |
+
running = "run" in text_prompt.lower()
|
| 44 |
+
jumping = "jump" in text_prompt.lower()
|
| 45 |
+
dancing = "danc" in text_prompt.lower()
|
| 46 |
+
turning = "turn" in text_prompt.lower() or "spin" in text_prompt.lower()
|
| 47 |
+
waving = "wave" in text_prompt.lower()
|
| 48 |
+
|
| 49 |
+
# Create a unique filename based on parameters
|
| 50 |
+
output_filename = f"output_{hash(text_prompt)}_{motion_length}_{seed}.mp4"
|
| 51 |
|
| 52 |
# Create a simple visualization script
|
| 53 |
+
with open("motion_animation.py", "w") as f:
|
| 54 |
+
f.write(f"""
|
| 55 |
import numpy as np
|
| 56 |
import matplotlib.pyplot as plt
|
| 57 |
from matplotlib.animation import FuncAnimation
|
| 58 |
import os
|
| 59 |
from mpl_toolkits.mplot3d import Axes3D
|
| 60 |
+
import random
|
| 61 |
|
| 62 |
+
# Set random seeds for reproducibility
|
| 63 |
+
np.random.seed({int(seed)})
|
| 64 |
+
random.seed({int(seed)})
|
| 65 |
+
|
| 66 |
+
# Create a motion based on text prompt
|
| 67 |
+
def generate_motion(frames={int(motion_length * 30)}):
|
| 68 |
joints = 16 # Number of joints in a simplified skeleton
|
| 69 |
dims = 3 # x, y, z
|
| 70 |
|
| 71 |
motion = np.zeros((frames, joints, dims))
|
| 72 |
|
| 73 |
+
# Parse animation parameters from text
|
| 74 |
+
walking = {walking}
|
| 75 |
+
running = {running}
|
| 76 |
+
jumping = {jumping}
|
| 77 |
+
dancing = {dancing}
|
| 78 |
+
turning = {turning}
|
| 79 |
+
waving = {waving}
|
| 80 |
+
|
| 81 |
+
# Set speed based on motion type
|
| 82 |
+
if running:
|
| 83 |
+
speed = 4.0
|
| 84 |
+
elif walking:
|
| 85 |
+
speed = 2.0
|
| 86 |
+
else:
|
| 87 |
+
speed = 1.0
|
| 88 |
+
|
| 89 |
+
# Create the motion
|
| 90 |
for frame in range(frames):
|
| 91 |
t = frame / frames
|
| 92 |
|
| 93 |
# Basic forward motion
|
| 94 |
+
if turning:
|
| 95 |
+
# Move in a circle
|
| 96 |
+
angle = t * 2 * np.pi * 2
|
| 97 |
+
motion[frame, :, 0] = np.cos(angle) * 2
|
| 98 |
+
motion[frame, :, 1] = np.sin(angle) * 2
|
| 99 |
+
else:
|
| 100 |
+
# Move forward
|
| 101 |
+
motion[frame, :, 0] = t * speed * 4
|
| 102 |
|
| 103 |
# Root joint (pelvis)
|
| 104 |
+
if jumping:
|
| 105 |
+
# Add jumping motion
|
| 106 |
+
jump_height = 0.5 + 0.5 * np.sin(t * 2 * np.pi * 3)
|
| 107 |
+
motion[frame, 0, 2] = jump_height
|
| 108 |
+
else:
|
| 109 |
+
# Regular walking bounce
|
| 110 |
+
bounce = 0.1 * np.sin(t * 2 * np.pi * speed * 2) if walking or running else 0.05
|
| 111 |
+
motion[frame, 0, 2] = bounce + 1
|
| 112 |
|
| 113 |
# Spine and head (joints 1, 2, 3)
|
| 114 |
for i in range(1, 4):
|
|
|
|
| 115 |
motion[frame, i, 2] = motion[frame, 0, 2] + i * 0.2 # Stack joints vertically
|
| 116 |
|
| 117 |
+
# Add dancing motion if needed
|
| 118 |
+
if dancing:
|
| 119 |
+
wiggle = 0.2 * np.sin(t * 2 * np.pi * 4 + np.pi * i/4)
|
| 120 |
+
motion[frame, 1:4, 1] = wiggle # Side-to-side motion for upper body
|
| 121 |
+
|
| 122 |
# Left leg (joints 4, 5, 6)
|
| 123 |
+
leg_freq = speed * 2 # Frequency of leg movement
|
| 124 |
+
swing_leg_l = np.sin(t * 2 * np.pi * leg_freq)
|
| 125 |
motion[frame, 4, 1] = 0.2 # Left hip position
|
| 126 |
motion[frame, 4, 2] = motion[frame, 0, 2] - 0.1
|
| 127 |
motion[frame, 5, 1] = 0.2
|
|
|
|
| 130 |
motion[frame, 6, 2] = motion[frame, 5, 2] - 0.5 + swing_leg_l * 0.3
|
| 131 |
|
| 132 |
# Right leg (joints 7, 8, 9)
|
| 133 |
+
swing_leg_r = np.sin(t * 2 * np.pi * leg_freq + np.pi) # Opposite phase
|
| 134 |
motion[frame, 7, 1] = -0.2 # Right hip position
|
| 135 |
motion[frame, 7, 2] = motion[frame, 0, 2] - 0.1
|
| 136 |
motion[frame, 8, 1] = -0.2
|
|
|
|
| 139 |
motion[frame, 9, 2] = motion[frame, 8, 2] - 0.5 + swing_leg_r * 0.3
|
| 140 |
|
| 141 |
# Left arm (joints 10, 11, 12)
|
| 142 |
+
if waving:
|
| 143 |
+
# Waving motion for left arm
|
| 144 |
+
if t > 0.3 and t < 0.7:
|
| 145 |
+
wave = 0.5 * np.sin(t * 2 * np.pi * 8)
|
| 146 |
+
motion[frame, 10, 1] = 0.3
|
| 147 |
+
motion[frame, 10, 2] = motion[frame, 3, 2] - 0.2
|
| 148 |
+
motion[frame, 11, 1] = 0.5
|
| 149 |
+
motion[frame, 11, 2] = motion[frame, 10, 2]
|
| 150 |
+
motion[frame, 12, 1] = 0.7
|
| 151 |
+
motion[frame, 12, 2] = motion[frame, 11, 2] + wave
|
| 152 |
+
else:
|
| 153 |
+
# Normal arm swing during non-waving
|
| 154 |
+
swing_arm_l = np.sin(t * 2 * np.pi * leg_freq + np.pi)
|
| 155 |
+
motion[frame, 10, 1] = 0.3
|
| 156 |
+
motion[frame, 10, 2] = motion[frame, 3, 2] - 0.2
|
| 157 |
+
motion[frame, 11, 1] = 0.3 + swing_arm_l * 0.2
|
| 158 |
+
motion[frame, 11, 2] = motion[frame, 10, 2] - 0.4
|
| 159 |
+
motion[frame, 12, 1] = 0.3 + swing_arm_l * 0.4
|
| 160 |
+
motion[frame, 12, 2] = motion[frame, 11, 2] - 0.4
|
| 161 |
+
else:
|
| 162 |
+
# Normal arm swing
|
| 163 |
+
swing_arm_l = np.sin(t * 2 * np.pi * leg_freq + np.pi)
|
| 164 |
+
motion[frame, 10, 1] = 0.3
|
| 165 |
+
motion[frame, 10, 2] = motion[frame, 3, 2] - 0.2
|
| 166 |
+
motion[frame, 11, 1] = 0.3 + swing_arm_l * 0.2
|
| 167 |
+
motion[frame, 11, 2] = motion[frame, 10, 2] - 0.4
|
| 168 |
+
motion[frame, 12, 1] = 0.3 + swing_arm_l * 0.4
|
| 169 |
+
motion[frame, 12, 2] = motion[frame, 11, 2] - 0.4
|
| 170 |
|
| 171 |
# Right arm (joints 13, 14, 15)
|
| 172 |
+
swing_arm_r = np.sin(t * 2 * np.pi * leg_freq)
|
| 173 |
+
motion[frame, 13, 1] = -0.3
|
| 174 |
motion[frame, 13, 2] = motion[frame, 3, 2] - 0.2
|
| 175 |
motion[frame, 14, 1] = -0.3 + swing_arm_r * 0.2
|
| 176 |
motion[frame, 14, 2] = motion[frame, 13, 2] - 0.4
|
|
|
|
| 181 |
|
| 182 |
def visualize_motion(output_path):
|
| 183 |
# Generate motion data
|
| 184 |
+
motion_data = generate_motion()
|
| 185 |
|
| 186 |
# Get dimensions
|
| 187 |
frames, joints, dims = motion_data.shape
|
|
|
|
| 203 |
ax.clear()
|
| 204 |
|
| 205 |
# Set axis limits
|
| 206 |
+
max_range = max(4, np.max(np.abs(motion_data)))
|
| 207 |
+
ax.set_xlim([-max_range/2, max_range/2 + motion_data[frame, 0, 0]])
|
| 208 |
+
ax.set_ylim([-max_range/2, max_range/2])
|
| 209 |
+
ax.set_zlim([0, max_range])
|
| 210 |
|
| 211 |
# Set labels
|
| 212 |
ax.set_xlabel('X (forward)')
|
|
|
|
| 224 |
[motion_data[frame, start, 1], motion_data[frame, end, 1]],
|
| 225 |
[motion_data[frame, start, 2], motion_data[frame, end, 2]], 'r-')
|
| 226 |
|
| 227 |
+
ax.set_title(f"{repr('{text_prompt}')} - Frame {{frame}}")
|
| 228 |
return ax
|
| 229 |
|
| 230 |
# Create animation
|
| 231 |
+
anim = FuncAnimation(fig, update, frames=min(180, motion_data.shape[0]), interval=1000/30)
|
| 232 |
|
| 233 |
# Save animation
|
| 234 |
os.makedirs(os.path.dirname(output_path) or '.', exist_ok=True)
|
| 235 |
anim.save(output_path, writer='ffmpeg', fps=30)
|
| 236 |
plt.close()
|
| 237 |
|
| 238 |
+
print(f"Animation saved to {{output_path}}")
|
| 239 |
return output_path
|
| 240 |
|
| 241 |
+
# Create and visualize a motion
|
| 242 |
+
visualize_motion("{output_filename}")
|
| 243 |
""")
|
| 244 |
|
| 245 |
# Run the script
|
| 246 |
+
subprocess.run(["python", "motion_animation.py"])
|
| 247 |
|
| 248 |
+
if os.path.exists(output_filename):
|
| 249 |
+
return output_filename
|
| 250 |
else:
|
| 251 |
return None
|
| 252 |
|
| 253 |
def text_to_motion(text_prompt, motion_length=3.0, seed=0):
|
| 254 |
"""Generate motion from text prompt using MDM"""
|
| 255 |
try:
|
| 256 |
+
# Inspect the MDM repository structure (limited output)
|
| 257 |
inspect_mdm_repo()
|
| 258 |
|
| 259 |
+
# Create a motion animation based on the input parameters
|
| 260 |
+
return create_motion_animation(text_prompt, motion_length, seed)
|
| 261 |
|
| 262 |
except Exception as e:
|
| 263 |
print(f"Error generating motion: {str(e)}")
|
|
|
|
| 274 |
],
|
| 275 |
outputs=gr.Video(label="Generated Motion"),
|
| 276 |
title="Motion Diffusion Model Demo",
|
| 277 |
+
description="Generate human motions from text descriptions. Try prompts with actions like 'walk', 'run', 'jump', 'dance', 'turn', or 'wave'."
|
| 278 |
)
|
| 279 |
|
| 280 |
# Launch the app
|