Update app.py
Browse files
app.py
CHANGED
|
@@ -8,20 +8,25 @@ import sys
|
|
| 8 |
import traceback
|
| 9 |
from huggingface_hub import hf_hub_download
|
| 10 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
# =========================================
|
| 12 |
# 1. Define Hugging Face dataset + weights
|
| 13 |
# =========================================
|
| 14 |
|
| 15 |
-
HF_DATASET_REPO = "roll-ai/FloVD-weights"
|
| 16 |
|
| 17 |
WEIGHT_FILES = {
|
| 18 |
-
"ckpt/FVSM/FloVD_FVSM_Controlnet.pt": "FVSM/
|
| 19 |
"ckpt/OMSM/selected_blocks.safetensors": "OMSM/selected_blocks.safetensors",
|
| 20 |
"ckpt/OMSM/pytorch_lora_weights.safetensors": "OMSM/pytorch_lora_weights.safetensors",
|
| 21 |
"ckpt/others/depth_anything_v2_metric_hypersim_vitb.pth": "others/depth_anything_v2_metric_hypersim_vitb.pth"
|
| 22 |
}
|
| 23 |
-
|
| 24 |
-
print("
|
|
|
|
| 25 |
def download_weights():
|
| 26 |
print("๐ Downloading model weights via huggingface_hub...")
|
| 27 |
for hf_path, local_rel_path in WEIGHT_FILES.items():
|
|
@@ -49,13 +54,14 @@ def print_ckpt_structure(base_path="ckpt"):
|
|
| 49 |
for f in files:
|
| 50 |
print(f"{sub_indent}๐ {f}", flush=True)
|
| 51 |
|
| 52 |
-
# Call it
|
| 53 |
print_ckpt_structure()
|
|
|
|
| 54 |
# =========================================
|
| 55 |
-
# 2. Import
|
| 56 |
# =========================================
|
| 57 |
|
| 58 |
from inference.flovd_demo import generate_video
|
|
|
|
| 59 |
def run_inference(prompt, image, pose_type, speed, use_flow_integration, cam_pose_name):
|
| 60 |
log_buffer = io.StringIO()
|
| 61 |
sys_stdout = sys.stdout
|
|
@@ -67,7 +73,6 @@ def run_inference(prompt, image, pose_type, speed, use_flow_integration, cam_pos
|
|
| 67 |
os.makedirs("input_images", exist_ok=True)
|
| 68 |
image_path = "input_images/input_image.png"
|
| 69 |
|
| 70 |
-
# โ
Convert NumPy to PIL if necessary
|
| 71 |
if not isinstance(image, Image.Image):
|
| 72 |
image = Image.fromarray(image.astype("uint8"))
|
| 73 |
|
|
@@ -110,11 +115,12 @@ def run_inference(prompt, image, pose_type, speed, use_flow_integration, cam_pos
|
|
| 110 |
|
| 111 |
return (video_path if video_path and os.path.exists(video_path) else None), logs
|
| 112 |
|
|
|
|
| 113 |
# =========================================
|
| 114 |
-
# 3. Gradio
|
| 115 |
# =========================================
|
| 116 |
|
| 117 |
-
|
| 118 |
fn=run_inference,
|
| 119 |
inputs=[
|
| 120 |
gr.Textbox(label="Prompt", value="A girl riding a bicycle through a park."),
|
|
@@ -137,8 +143,48 @@ demo = gr.Interface(
|
|
| 137 |
description="Upload an image and prompt to generate motion-controlled video using FloVD and CogVideoX."
|
| 138 |
)
|
| 139 |
|
| 140 |
-
#
|
| 141 |
-
#
|
| 142 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 143 |
if __name__ == "__main__":
|
| 144 |
-
|
|
|
|
| 8 |
import traceback
|
| 9 |
from huggingface_hub import hf_hub_download
|
| 10 |
|
| 11 |
+
# For live system monitoring
|
| 12 |
+
import psutil
|
| 13 |
+
import GPUtil
|
| 14 |
+
|
| 15 |
# =========================================
|
| 16 |
# 1. Define Hugging Face dataset + weights
|
| 17 |
# =========================================
|
| 18 |
|
| 19 |
+
HF_DATASET_REPO = "roll-ai/FloVD-weights"
|
| 20 |
|
| 21 |
WEIGHT_FILES = {
|
| 22 |
+
"ckpt/FVSM/FloVD_FVSM_Controlnet.pt": "FVSM/FloVD_FVSM_Controlnet.pt",
|
| 23 |
"ckpt/OMSM/selected_blocks.safetensors": "OMSM/selected_blocks.safetensors",
|
| 24 |
"ckpt/OMSM/pytorch_lora_weights.safetensors": "OMSM/pytorch_lora_weights.safetensors",
|
| 25 |
"ckpt/others/depth_anything_v2_metric_hypersim_vitb.pth": "others/depth_anything_v2_metric_hypersim_vitb.pth"
|
| 26 |
}
|
| 27 |
+
|
| 28 |
+
print("\nDownloading model...", flush=True)
|
| 29 |
+
|
| 30 |
def download_weights():
|
| 31 |
print("๐ Downloading model weights via huggingface_hub...")
|
| 32 |
for hf_path, local_rel_path in WEIGHT_FILES.items():
|
|
|
|
| 54 |
for f in files:
|
| 55 |
print(f"{sub_indent}๐ {f}", flush=True)
|
| 56 |
|
|
|
|
| 57 |
print_ckpt_structure()
|
| 58 |
+
|
| 59 |
# =========================================
|
| 60 |
+
# 2. Import FloVD generation pipeline
|
| 61 |
# =========================================
|
| 62 |
|
| 63 |
from inference.flovd_demo import generate_video
|
| 64 |
+
|
| 65 |
def run_inference(prompt, image, pose_type, speed, use_flow_integration, cam_pose_name):
|
| 66 |
log_buffer = io.StringIO()
|
| 67 |
sys_stdout = sys.stdout
|
|
|
|
| 73 |
os.makedirs("input_images", exist_ok=True)
|
| 74 |
image_path = "input_images/input_image.png"
|
| 75 |
|
|
|
|
| 76 |
if not isinstance(image, Image.Image):
|
| 77 |
image = Image.fromarray(image.astype("uint8"))
|
| 78 |
|
|
|
|
| 115 |
|
| 116 |
return (video_path if video_path and os.path.exists(video_path) else None), logs
|
| 117 |
|
| 118 |
+
|
| 119 |
# =========================================
|
| 120 |
+
# 3. Define FloVD Gradio Interface
|
| 121 |
# =========================================
|
| 122 |
|
| 123 |
+
video_interface = gr.Interface(
|
| 124 |
fn=run_inference,
|
| 125 |
inputs=[
|
| 126 |
gr.Textbox(label="Prompt", value="A girl riding a bicycle through a park."),
|
|
|
|
| 143 |
description="Upload an image and prompt to generate motion-controlled video using FloVD and CogVideoX."
|
| 144 |
)
|
| 145 |
|
| 146 |
+
# =========================================
|
| 147 |
+
# 4. Live System Monitor
|
| 148 |
+
# =========================================
|
| 149 |
+
|
| 150 |
+
def get_system_stats():
|
| 151 |
+
cpu = psutil.cpu_percent()
|
| 152 |
+
mem = psutil.virtual_memory()
|
| 153 |
+
disk = psutil.disk_usage('/')
|
| 154 |
+
try:
|
| 155 |
+
gpus = GPUtil.getGPUs()
|
| 156 |
+
gpu_info = "\n".join([
|
| 157 |
+
f"GPU {i}: {gpu.name}, {gpu.memoryUsed}MB / {gpu.memoryTotal}MB, Util: {gpu.load * 100:.1f}%"
|
| 158 |
+
for i, gpu in enumerate(gpus)
|
| 159 |
+
]) if gpus else "No GPU detected"
|
| 160 |
+
except Exception as e:
|
| 161 |
+
gpu_info = f"GPU info error: {e}"
|
| 162 |
+
|
| 163 |
+
return (
|
| 164 |
+
f"๐ง CPU Usage: {cpu}%\n"
|
| 165 |
+
f"๐พ RAM: {mem.used / 1e9:.2f} GB / {mem.total / 1e9:.2f} GB ({mem.percent}%)\n"
|
| 166 |
+
f"๐๏ธ Disk: {disk.used / 1e9:.2f} GB / {disk.total / 1e9:.2f} GB ({disk.percent}%)\n"
|
| 167 |
+
f"๐ฎ {gpu_info}"
|
| 168 |
+
)
|
| 169 |
+
|
| 170 |
+
with gr.Blocks() as monitor_tab:
|
| 171 |
+
gr.Markdown("## ๐ Live System Resource Monitor")
|
| 172 |
+
stats_box = gr.Textbox(label="Live Stats", lines=10, interactive=False)
|
| 173 |
+
gr.Live(stats_box.update, fn=get_system_stats, every=2.0)
|
| 174 |
+
|
| 175 |
+
# =========================================
|
| 176 |
+
# 5. Combine Tabs: FloVD + Monitor
|
| 177 |
+
# =========================================
|
| 178 |
+
|
| 179 |
+
with gr.Blocks() as app:
|
| 180 |
+
with gr.Tab("๐ฅ Video Generator"):
|
| 181 |
+
video_interface.render()
|
| 182 |
+
with gr.Tab("๐ System Monitor"):
|
| 183 |
+
monitor_tab.render()
|
| 184 |
+
|
| 185 |
+
# =========================================
|
| 186 |
+
# 6. Launch App
|
| 187 |
+
# =========================================
|
| 188 |
+
|
| 189 |
if __name__ == "__main__":
|
| 190 |
+
app.launch(server_name="0.0.0.0", server_port=7860, debug=True, show_error=True)
|