Spaces:
Running on Zero
Running on Zero
File size: 4,907 Bytes
42cb75b 49032ca 42cb75b c6fccb0 42cb75b b684910 42cb75b 49032ca 42cb75b 49032ca 42cb75b 49032ca 42cb75b 49032ca 42cb75b 49032ca 42cb75b 49032ca 42cb75b 49032ca 42cb75b | 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 | import os
import shutil
import subprocess
import sys
import tempfile
import traceback
from pathlib import Path
import gradio as gr
import numpy as np
from PIL import Image
import spaces
MODEL_REPO = "facebook/sam-3d-objects"
CHECKPOINT_DIR = Path("checkpoints/hf")
SOURCE_DIR = Path("sam-3d-objects")
def run(cmd, cwd=None):
proc = subprocess.run(
cmd,
cwd=cwd,
text=True,
stdout=subprocess.PIPE,
stderr=subprocess.STDOUT,
check=False,
)
if proc.returncode != 0:
raise RuntimeError(proc.stdout)
return proc.stdout
def diagnostic_text():
lines = []
lines.append(f"Python: {sys.version.split()[0]}")
lines.append(f"HF_TOKEN set: {'yes' if os.getenv('HF_TOKEN') else 'no'}")
try:
import torch
lines.append(f"Torch: {torch.__version__}")
lines.append(f"CUDA available: {torch.cuda.is_available()}")
if torch.cuda.is_available():
props = torch.cuda.get_device_properties(0)
vram_gb = props.total_memory / (1024**3)
lines.append(f"GPU: {props.name} ({vram_gb:.1f} GB VRAM)")
except Exception as exc:
lines.append(f"Torch import failed: {exc}")
lines.append(f"Source present: {SOURCE_DIR.exists()}")
lines.append(f"Checkpoints present: {(CHECKPOINT_DIR / 'pipeline.yaml').exists()}")
return "\n".join(lines)
def ensure_repo():
if SOURCE_DIR.exists():
return
run(["git", "clone", "--depth", "1", "https://github.com/facebookresearch/sam-3d-objects.git", str(SOURCE_DIR)])
def ensure_checkpoints():
if (CHECKPOINT_DIR / "pipeline.yaml").exists():
return
from huggingface_hub import snapshot_download
token = os.getenv("HF_TOKEN")
if not token:
raise RuntimeError("HF_TOKEN secret is not set. Add a token with access to facebook/sam-3d-objects.")
tmp = snapshot_download(
repo_id=MODEL_REPO,
repo_type="model",
token=token,
local_dir="checkpoints/hf-download",
max_workers=1,
)
nested = Path(tmp) / "checkpoints"
CHECKPOINT_DIR.parent.mkdir(parents=True, exist_ok=True)
if CHECKPOINT_DIR.exists():
shutil.rmtree(CHECKPOINT_DIR)
shutil.move(str(nested), str(CHECKPOINT_DIR))
def ensure_runtime():
ensure_repo()
if str(SOURCE_DIR / "notebook") not in sys.path:
sys.path.append(str(SOURCE_DIR / "notebook"))
ensure_checkpoints()
_inference = None
def get_inference():
global _inference
if _inference is not None:
return _inference
ensure_runtime()
from inference import Inference
_inference = Inference(str(CHECKPOINT_DIR / "pipeline.yaml"), compile=False)
return _inference
def prepare_mask(mask_image):
if mask_image is None:
raise gr.Error("Provide a binary mask image. White pixels should mark the object.")
mask = Image.fromarray(mask_image).convert("L")
mask = mask.point(lambda value: 255 if value > 127 else 0)
return mask
@spaces.GPU(duration=120)
def reconstruct(image, mask_image, seed):
try:
if image is None:
raise gr.Error("Upload an input image.")
with tempfile.TemporaryDirectory() as tmpdir:
tmp = Path(tmpdir)
image_path = tmp / "image.png"
mask_path = tmp / "mask.png"
output_path = tmp / "sam3d-output.ply"
Image.fromarray(image).convert("RGB").save(image_path)
prepare_mask(mask_image).save(mask_path)
inference = get_inference()
from inference import load_image, load_mask
loaded_image = load_image(str(image_path))
loaded_mask = load_mask(str(mask_path))
output = inference(loaded_image, loaded_mask, seed=int(seed))
output["gs"].save_ply(str(output_path))
final_path = Path("outputs") / "sam3d-output.ply"
final_path.parent.mkdir(exist_ok=True)
shutil.copyfile(output_path, final_path)
return str(final_path), diagnostic_text()
except Exception:
return None, diagnostic_text() + "\n\n" + traceback.format_exc()
with gr.Blocks(title="SAM 3D Objects") as demo:
gr.Markdown("# SAM 3D Objects")
gr.Markdown("Upload an image and an object mask. White mask pixels are reconstructed.")
with gr.Row():
image = gr.Image(label="Image", type="numpy")
mask = gr.Image(label="Object mask", type="numpy", image_mode="L")
seed = gr.Number(label="Seed", value=42, precision=0)
run_button = gr.Button("Reconstruct")
model_output = gr.File(label="Gaussian splat PLY")
status = gr.Textbox(label="Diagnostics", lines=8, value=diagnostic_text)
run_button.click(reconstruct, inputs=[image, mask, seed], outputs=[model_output, status], api_name="reconstruct")
if __name__ == "__main__":
demo.queue(max_size=4).launch()
|