f3cb18ac4d
Bundle all per-image modalities (depth, edge, chm, segm) into a single .safetensors file for fast training DataLoader reads (~0.1ms zero-copy mmap vs ~5ms for 4x PNG). Adds consolidate stage after main pipeline stages, save_safetensors/cleanup_npy config flags, resume support, and 10 new tests. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
553 lines
20 KiB
Python
553 lines
20 KiB
Python
"""Entry point for the depth/edges/segmentation/chmv2 augmentation pipeline.
|
|
|
|
All parameters loaded from gin config files — no argparse.
|
|
Sequential stage processing: one model at a time, load → process all → unload.
|
|
|
|
Usage:
|
|
python -m src.main
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import gc
|
|
import json
|
|
import logging
|
|
import time
|
|
from datetime import datetime
|
|
from pathlib import Path
|
|
from typing import Any
|
|
|
|
import numpy as np
|
|
import torch
|
|
from torch.utils.data import DataLoader
|
|
from tqdm import tqdm
|
|
|
|
from src.conf.config_loader import load_all_configs
|
|
from src.conf.pipeline_conf import PipelineConfig
|
|
from src.conf.hardware_conf import HardwareConfig
|
|
from src.conf.models_conf import ModelsConfig
|
|
from src.conf.input_conf import InputConfig
|
|
from src.conf.seg_conf import SegConfig
|
|
from src.utils.utils_file_dir import get_proj_dir
|
|
|
|
from src.augmentor.dataset import (
|
|
AugmentDataset, ImageRecord, attach_output_dirs,
|
|
discover_images, filter_completed, split_by_view,
|
|
)
|
|
from src.augmentor.inference import (
|
|
compute_edges_from_depth, infer_depth_batch, infer_chmv2_batch,
|
|
infer_segmentation_batch,
|
|
)
|
|
from src.augmentor.io_utils import (
|
|
save_depth_async, save_chmv2_async, save_edges_async,
|
|
save_segmentation_async, consolidate_safetensors,
|
|
setup_logging, shutdown_io_pool,
|
|
)
|
|
from src.augmentor.models import (
|
|
load_depth_model, load_chmv2_model, load_segmentation_model, unload_model,
|
|
)
|
|
from src.utils.profiler import (
|
|
log_system_info, log_disk_info, log_vram_snapshot, log_ram_snapshot,
|
|
)
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
_STAGE_EMOJI = {
|
|
"depth": "🌊",
|
|
"edges": "🔪",
|
|
"segmentation": "🗺️",
|
|
"chmv2": "🦕",
|
|
"concat": "🧩",
|
|
"consolidate": "📦",
|
|
}
|
|
|
|
|
|
def _silence_model_loggers() -> None:
|
|
"""Suppress verbose inference logs from all models."""
|
|
import os
|
|
import warnings
|
|
|
|
os.environ["DA3_LOG_LEVEL"] = "ERROR"
|
|
os.environ["TRANSFORMERS_VERBOSITY"] = "error"
|
|
os.environ["HF_HUB_DISABLE_PROGRESS_BARS"] = "1"
|
|
|
|
for name in (
|
|
"depth_anything_3", "depth_anything_3.api",
|
|
"depth_anything_3.utils.logger",
|
|
"transformers", "transformers.modeling_utils",
|
|
"transformers.configuration_utils", "transformers.image_processing_utils",
|
|
"transformers.image_processing_base",
|
|
"sam3", "segearthov3_segmentor",
|
|
"huggingface_hub", "httpx", "filelock",
|
|
"numexpr", "numexpr.utils",
|
|
"torch", "py.warnings",
|
|
):
|
|
logging.getLogger(name).setLevel(logging.ERROR)
|
|
|
|
warnings.filterwarnings("ignore", category=DeprecationWarning)
|
|
warnings.filterwarnings("ignore", category=FutureWarning)
|
|
warnings.filterwarnings("ignore", message=".*not sharded.*")
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Stage runners
|
|
# ---------------------------------------------------------------------------
|
|
|
|
def _resolve_image_sizes(
|
|
records: list[ImageRecord],
|
|
input_conf: InputConfig,
|
|
) -> list[tuple[list[ImageRecord], int, str]]:
|
|
"""Split records into groups by target resolution.
|
|
|
|
Returns list of (records, image_size, label) tuples. When
|
|
``query_image_size == image_size`` a single group is returned (no split).
|
|
"""
|
|
if input_conf.query_image_size == input_conf.image_size:
|
|
return [(records, input_conf.image_size, "all")]
|
|
db_recs, query_recs = split_by_view(records)
|
|
groups: list[tuple[list[ImageRecord], int, str]] = []
|
|
if db_recs:
|
|
groups.append((db_recs, input_conf.image_size, f"db {input_conf.image_size}"))
|
|
if query_recs:
|
|
groups.append((query_recs, input_conf.query_image_size, f"query {input_conf.query_image_size}"))
|
|
return groups
|
|
|
|
|
|
def run_depth_stage(
|
|
records: list[ImageRecord],
|
|
pipeline_conf: PipelineConfig,
|
|
hw_conf: HardwareConfig,
|
|
models_conf: ModelsConfig,
|
|
input_conf: InputConfig,
|
|
device: torch.device,
|
|
) -> None:
|
|
"""🌊 Load DA3, process all images, unload."""
|
|
model = load_depth_model(models_conf, hw_conf, device)
|
|
|
|
for group_records, sz, label in _resolve_image_sizes(records, input_conf):
|
|
bs = hw_conf.find_batch_size(
|
|
inference_fn=lambda x: infer_depth_batch(model, x, device),
|
|
input_shape=(3, sz, sz),
|
|
device=device,
|
|
)
|
|
|
|
ds = AugmentDataset(group_records, image_size=sz)
|
|
logger.info("🌊 [depth/%s] DataLoader: batch_size=%d, %d images, %d batches",
|
|
label, bs, len(ds), (len(ds) + bs - 1) // bs)
|
|
loader = DataLoader(
|
|
ds, batch_size=bs, shuffle=False,
|
|
num_workers=hw_conf.num_workers, pin_memory=True,
|
|
persistent_workers=hw_conf.num_workers > 0,
|
|
prefetch_factor=4 if hw_conf.num_workers > 0 else None,
|
|
)
|
|
|
|
total_images = len(ds)
|
|
pbar = tqdm(loader, desc=f"🌊 depth/{label} (bs={bs})", unit="batch",
|
|
total=len(loader), colour="green",
|
|
bar_format="{l_bar}{bar}| {n_fmt}/{total_fmt} batches [{elapsed}<{remaining}, {rate_fmt}]")
|
|
processed = 0
|
|
for batch in pbar:
|
|
depths = infer_depth_batch(model, batch["image_raw"], device)
|
|
for i in range(depths.shape[0]):
|
|
save_depth_async(depths[i], Path(batch["output_dir"][i]),
|
|
stem=batch["stem"][i],
|
|
save_npy=pipeline_conf.save_npy,
|
|
save_vis=pipeline_conf.save_vis)
|
|
processed += depths.shape[0]
|
|
pbar.set_postfix(images=f"{processed}/{total_images}")
|
|
|
|
shutdown_io_pool()
|
|
|
|
unload_model(model)
|
|
|
|
|
|
def run_edges_stage(
|
|
records: list[ImageRecord],
|
|
pipeline_conf: PipelineConfig,
|
|
batch_size: int = 32,
|
|
) -> None:
|
|
"""🔪 Compute Sobel edges from saved depth (CPU, batched)."""
|
|
valid: list[ImageRecord] = []
|
|
for r in records:
|
|
depth_png = r.output_dir / f"{r.stem}_depth.png"
|
|
depth_npy = r.output_dir / f"{r.stem}_depth.npy"
|
|
if depth_png.exists() or depth_npy.exists():
|
|
valid.append(r)
|
|
else:
|
|
logger.warning("⚠️ No depth for %s, skipping edges.", r.rel_path)
|
|
|
|
total_images = len(valid)
|
|
pbar = tqdm(range(0, len(valid), batch_size), desc="🔪 edges (sobel)", unit="batch",
|
|
colour="cyan",
|
|
bar_format="{l_bar}{bar}| {n_fmt}/{total_fmt} batches [{elapsed}<{remaining}, {rate_fmt}]")
|
|
processed = 0
|
|
for start in pbar:
|
|
chunk = valid[start : start + batch_size]
|
|
depth_tensors = []
|
|
for r in chunk:
|
|
npy_path = r.output_dir / f"{r.stem}_depth.npy"
|
|
png_path = r.output_dir / f"{r.stem}_depth.png"
|
|
if npy_path.exists():
|
|
d = np.load(npy_path).astype(np.float32)
|
|
else:
|
|
from PIL import Image
|
|
d = np.array(Image.open(png_path)).astype(np.float32) / 255.0
|
|
if d.ndim == 2:
|
|
d = d[np.newaxis]
|
|
depth_tensors.append(torch.from_numpy(d))
|
|
depths = torch.stack(depth_tensors)
|
|
if depths.ndim == 3:
|
|
depths = depths.unsqueeze(1)
|
|
edges_batch = compute_edges_from_depth(depths)
|
|
for j, r in enumerate(chunk):
|
|
save_edges_async(edges_batch[j], r.output_dir, stem=r.stem,
|
|
save_npy=pipeline_conf.save_npy,
|
|
save_vis=pipeline_conf.save_vis)
|
|
processed += len(chunk)
|
|
pbar.set_postfix(images=f"{processed}/{total_images}")
|
|
|
|
shutdown_io_pool()
|
|
|
|
|
|
def run_chmv2_stage(
|
|
records: list[ImageRecord],
|
|
pipeline_conf: PipelineConfig,
|
|
hw_conf: HardwareConfig,
|
|
models_conf: ModelsConfig,
|
|
input_conf: InputConfig,
|
|
device: torch.device,
|
|
) -> None:
|
|
"""🦕 Load CHMv2 (DINOv3), process all images, unload."""
|
|
model, processor = load_chmv2_model(models_conf, hw_conf, device)
|
|
|
|
for group_records, sz, label in _resolve_image_sizes(records, input_conf):
|
|
bs = hw_conf.find_batch_size(
|
|
inference_fn=lambda x: infer_chmv2_batch(model, processor, x, device),
|
|
input_shape=(3, sz, sz),
|
|
device=device,
|
|
)
|
|
|
|
ds = AugmentDataset(group_records, image_size=sz)
|
|
logger.info("🦕 [chmv2/%s] DataLoader: batch_size=%d, %d images, %d batches",
|
|
label, bs, len(ds), (len(ds) + bs - 1) // bs)
|
|
loader = DataLoader(
|
|
ds, batch_size=bs, shuffle=False,
|
|
num_workers=hw_conf.num_workers, pin_memory=True,
|
|
persistent_workers=hw_conf.num_workers > 0,
|
|
prefetch_factor=4 if hw_conf.num_workers > 0 else None,
|
|
)
|
|
|
|
total_images = len(ds)
|
|
pbar = tqdm(loader, desc=f"🦕 chmv2/{label} (bs={bs})", unit="batch",
|
|
total=len(loader), colour="blue",
|
|
bar_format="{l_bar}{bar}| {n_fmt}/{total_fmt} batches [{elapsed}<{remaining}, {rate_fmt}]")
|
|
processed = 0
|
|
for batch in pbar:
|
|
depths = infer_chmv2_batch(model, processor, batch["image_raw"], device)
|
|
for i in range(depths.shape[0]):
|
|
save_chmv2_async(depths[i], Path(batch["output_dir"][i]),
|
|
stem=batch["stem"][i],
|
|
save_npy=pipeline_conf.save_npy,
|
|
save_vis=pipeline_conf.save_vis)
|
|
processed += depths.shape[0]
|
|
pbar.set_postfix(images=f"{processed}/{total_images}")
|
|
|
|
shutdown_io_pool()
|
|
|
|
unload_model(model)
|
|
|
|
|
|
def run_segmentation_stage(
|
|
records: list[ImageRecord],
|
|
pipeline_conf: PipelineConfig,
|
|
hw_conf: HardwareConfig,
|
|
models_conf: ModelsConfig,
|
|
input_conf: InputConfig,
|
|
seg_conf: SegConfig,
|
|
device: torch.device,
|
|
) -> None:
|
|
"""🗺️ Load segmentation model, process all images, unload."""
|
|
model, seg_config = load_segmentation_model(models_conf, hw_conf, seg_conf, device)
|
|
num_classes = seg_config.get("num_classes", 150)
|
|
is_segearth = seg_config.get("type") == "segearth-ov3"
|
|
|
|
for group_records, sz, label in _resolve_image_sizes(records, input_conf):
|
|
if is_segearth:
|
|
bs = 16
|
|
else:
|
|
bs = hw_conf.find_batch_size(
|
|
inference_fn=lambda x: infer_segmentation_batch(model, seg_config, x, device),
|
|
input_shape=(3, sz, sz),
|
|
device=device,
|
|
)
|
|
|
|
ds = AugmentDataset(group_records, image_size=sz)
|
|
total_images = len(ds)
|
|
logger.info("🗺️ [segmentation/%s] DataLoader: batch_size=%d, %d images, %d batches",
|
|
label, bs, total_images, (total_images + bs - 1) // bs)
|
|
loader = DataLoader(
|
|
ds, batch_size=bs, shuffle=False,
|
|
num_workers=hw_conf.num_workers, pin_memory=True,
|
|
persistent_workers=hw_conf.num_workers > 0,
|
|
prefetch_factor=4 if hw_conf.num_workers > 0 else None,
|
|
)
|
|
|
|
seg_type = "SegEarth-OV3" if is_segearth else "SegFormer"
|
|
pbar = tqdm(loader, desc=f"🗺️ seg/{label} {seg_type} (bs={bs})", unit="batch",
|
|
colour="yellow",
|
|
bar_format="{l_bar}{bar}| {n_fmt}/{total_fmt} batches [{elapsed}<{remaining}, {rate_fmt}]")
|
|
processed = 0
|
|
for batch in pbar:
|
|
segs = infer_segmentation_batch(
|
|
model, seg_config, batch["image_raw"], device,
|
|
)
|
|
for j in range(segs.shape[0]):
|
|
save_segmentation_async(
|
|
segs[j], Path(batch["output_dir"][j]), stem=batch["stem"][j],
|
|
save_npy=pipeline_conf.save_npy, save_vis=pipeline_conf.save_vis,
|
|
num_classes=num_classes,
|
|
)
|
|
processed += segs.shape[0]
|
|
pbar.set_postfix(images=f"{processed}/{total_images}")
|
|
|
|
shutdown_io_pool()
|
|
|
|
unload_model(model)
|
|
|
|
|
|
def run_consolidate_stage(
|
|
records: list[ImageRecord],
|
|
cleanup_npy: bool = False,
|
|
) -> None:
|
|
"""📦 Bundle per-image .npy/.png modalities into .safetensors files."""
|
|
total = len(records)
|
|
written = 0
|
|
pbar = tqdm(records, desc="📦 consolidate → safetensors", unit="img",
|
|
colour="magenta",
|
|
bar_format="{l_bar}{bar}| {n_fmt}/{total_fmt} [{elapsed}<{remaining}, {rate_fmt}]")
|
|
for r in pbar:
|
|
ok = consolidate_safetensors(r.output_dir, r.stem, cleanup_npy=cleanup_npy)
|
|
if ok:
|
|
written += 1
|
|
pbar.set_postfix(written=f"{written}/{total}")
|
|
logger.info("📦 Consolidated %d / %d images to .safetensors.", written, total)
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Pipeline orchestration
|
|
# ---------------------------------------------------------------------------
|
|
|
|
def run_pipeline(
|
|
pipeline_conf: PipelineConfig,
|
|
hw_conf: HardwareConfig,
|
|
models_conf: ModelsConfig,
|
|
input_conf: InputConfig,
|
|
seg_conf: SegConfig,
|
|
) -> None:
|
|
"""Execute the full augmentation pipeline: one stage at a time."""
|
|
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
|
if device.type != "cuda":
|
|
logger.warning("⚠️ CUDA not available, running on CPU (very slow).")
|
|
|
|
_silence_model_loggers()
|
|
|
|
# System profiling at startup.
|
|
log_system_info()
|
|
log_disk_info(Path(pipeline_conf.input_root), Path(pipeline_conf.output_root))
|
|
|
|
# Discover images.
|
|
input_root = Path(pipeline_conf.input_root)
|
|
output_root = Path(pipeline_conf.output_root)
|
|
logger.info(
|
|
"🔍 Discovering images in %s (subset=%s, source=%s) ...",
|
|
input_root, pipeline_conf.subset or "all", pipeline_conf.source or "all",
|
|
)
|
|
all_records = discover_images(input_root, subset=pipeline_conf.subset,
|
|
source=pipeline_conf.source)
|
|
all_records = attach_output_dirs(all_records, output_root)
|
|
logger.info("📸 Found %d images.", len(all_records))
|
|
|
|
if not all_records:
|
|
logger.error("❌ No images found. Check input_root in pipeline.gin.")
|
|
return
|
|
|
|
# Pre-create all output directories in one pass.
|
|
logger.info("📁 Pre-creating output directories...")
|
|
seen_dirs: set[str] = set()
|
|
for r in all_records:
|
|
d = str(r.output_dir)
|
|
if d not in seen_dirs:
|
|
r.output_dir.mkdir(parents=True, exist_ok=True)
|
|
seen_dirs.add(d)
|
|
|
|
# Process each stage sequentially.
|
|
stage_times: dict[str, float] = {}
|
|
stage_counts: dict[str, int] = {}
|
|
failed_stages: set[str] = set()
|
|
|
|
for stage in pipeline_conf.stages:
|
|
emoji = _STAGE_EMOJI.get(stage, "⚙️")
|
|
|
|
if stage == "edges" and "depth" in failed_stages:
|
|
logger.error("❌ [edges] Skipped — depth stage failed.")
|
|
failed_stages.add(stage)
|
|
continue
|
|
|
|
pending, skipped = filter_completed(all_records, stage)
|
|
|
|
logger.info("%s [%s] %d pending, %d skipped.", emoji, stage, len(pending), skipped)
|
|
stage_counts[stage] = len(pending)
|
|
|
|
if not pending:
|
|
stage_times[stage] = 0.0
|
|
continue
|
|
|
|
logger.info("%s [%s] Starting stage...", emoji, stage)
|
|
log_vram_snapshot(f"before {stage}")
|
|
log_ram_snapshot(f"before {stage}")
|
|
t0 = time.perf_counter()
|
|
try:
|
|
if stage == "depth":
|
|
run_depth_stage(pending, pipeline_conf, hw_conf, models_conf,
|
|
input_conf, device)
|
|
elif stage == "edges":
|
|
run_edges_stage(pending, pipeline_conf)
|
|
elif stage == "segmentation":
|
|
run_segmentation_stage(pending, pipeline_conf, hw_conf, models_conf,
|
|
input_conf, seg_conf, device)
|
|
elif stage == "chmv2":
|
|
run_chmv2_stage(pending, pipeline_conf, hw_conf, models_conf,
|
|
input_conf, device)
|
|
except Exception:
|
|
logger.exception("💥 Stage '%s' failed.", stage)
|
|
failed_stages.add(stage)
|
|
elapsed = time.perf_counter() - t0
|
|
stage_times[stage] = elapsed
|
|
if stage not in failed_stages:
|
|
logger.info("✅ [%s] Completed in %.1f s (%d images).", stage, elapsed, len(pending))
|
|
log_vram_snapshot(f"after {stage}")
|
|
log_ram_snapshot(f"after {stage}")
|
|
|
|
# SafeTensors consolidation: bundle all modalities per image.
|
|
if pipeline_conf.save_safetensors:
|
|
pending_st, skipped_st = filter_completed(all_records, "consolidate")
|
|
logger.info("📦 [consolidate] %d pending, %d skipped.", len(pending_st), skipped_st)
|
|
if pending_st:
|
|
t0 = time.perf_counter()
|
|
run_consolidate_stage(pending_st, cleanup_npy=pipeline_conf.cleanup_npy)
|
|
elapsed_st = time.perf_counter() - t0
|
|
stage_times["consolidate"] = elapsed_st
|
|
stage_counts["consolidate"] = len(pending_st)
|
|
logger.info("✅ [consolidate] Completed in %.1f s (%d images).",
|
|
elapsed_st, len(pending_st))
|
|
else:
|
|
stage_times["consolidate"] = 0.0
|
|
stage_counts["consolidate"] = 0
|
|
|
|
# Manifest.
|
|
manifest = {
|
|
"pipeline_version": "3.3.0-safetensors",
|
|
"image_size_db": input_conf.image_size,
|
|
"image_size_query": input_conf.query_image_size,
|
|
"profile": hw_conf.profile_name,
|
|
"models": {
|
|
"depth": models_conf.depth_model_id,
|
|
"edges": "Sobel from depth (CPU)",
|
|
"segmentation": models_conf.seg_model_type,
|
|
"chmv2": models_conf.chmv2_model_id,
|
|
},
|
|
"save_safetensors": pipeline_conf.save_safetensors,
|
|
"seg_prompts": seg_conf.prompts,
|
|
"total_images": len(all_records),
|
|
"stages": {
|
|
s: {"processed": stage_counts.get(s, 0),
|
|
"time_sec": round(stage_times.get(s, 0), 1)}
|
|
for s in pipeline_conf.stages
|
|
},
|
|
"timestamp": datetime.now().isoformat(),
|
|
}
|
|
manifest_path = output_root / "manifest.json"
|
|
manifest_path.parent.mkdir(parents=True, exist_ok=True)
|
|
manifest_path.write_text(
|
|
json.dumps(manifest, indent=2, ensure_ascii=False), encoding="utf-8",
|
|
)
|
|
|
|
# Summary.
|
|
total = sum(stage_times.values())
|
|
logger.info("=" * 60)
|
|
logger.info("🏁 DONE: %d images, %.1f s total", len(all_records), total)
|
|
for s, t in stage_times.items():
|
|
cnt = stage_counts.get(s, len(all_records))
|
|
fps = cnt / t if t > 0 else 0
|
|
emoji = _STAGE_EMOJI.get(s, "⚙️")
|
|
logger.info(" %s %-13s %6.1f s (%d images, %.0f FPS)", emoji, s, t, cnt, fps)
|
|
logger.info("📂 Output: %s", output_root)
|
|
logger.info("=" * 60)
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Entry point
|
|
# ---------------------------------------------------------------------------
|
|
|
|
def main() -> None:
|
|
"""Load all gin configs and run the augmentation pipeline.
|
|
|
|
Supports CLI gin overrides for quick mode switches::
|
|
|
|
# Process only query (drone) images:
|
|
python -m src.main --gin "PipelineConfig.source = 'query'"
|
|
|
|
# Process only db (satellite) images:
|
|
python -m src.main --gin "PipelineConfig.source = 'db'"
|
|
"""
|
|
import argparse
|
|
|
|
parser = argparse.ArgumentParser(description="Augmentation pipeline")
|
|
parser.add_argument("--gin", action="append", default=[],
|
|
help="Gin parameter overrides (repeatable)")
|
|
args = parser.parse_args()
|
|
|
|
_silence_model_loggers()
|
|
|
|
proj_dir = get_proj_dir()
|
|
path2cfg = f"{proj_dir}in/config_files/"
|
|
|
|
# Load configs with optional CLI overrides.
|
|
if args.gin:
|
|
import gin as _gin
|
|
cfg_dir = Path(path2cfg)
|
|
gin_files = sorted(cfg_dir.glob("*.gin"))
|
|
_gin.clear_config()
|
|
_gin.parse_config_files_and_bindings(
|
|
config_files=[str(f) for f in gin_files],
|
|
bindings=args.gin,
|
|
)
|
|
configs = {
|
|
"pipeline": PipelineConfig(),
|
|
"hardware": HardwareConfig(),
|
|
"models": ModelsConfig(),
|
|
"input": InputConfig(),
|
|
"seg": SegConfig(),
|
|
}
|
|
else:
|
|
configs = load_all_configs(path2cfg)
|
|
|
|
pipeline_conf: PipelineConfig = configs["pipeline"]
|
|
setup_logging(pipeline_conf.log_level,
|
|
log_file=Path(pipeline_conf.output_root) / "pipeline.log")
|
|
|
|
torch.manual_seed(42)
|
|
np.random.seed(42)
|
|
|
|
run_pipeline(
|
|
configs["pipeline"],
|
|
configs["hardware"],
|
|
configs["models"],
|
|
configs["input"],
|
|
configs["seg"],
|
|
)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|