Datasets:
File size: 6,381 Bytes
5d49765 | 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 | """Local verification of the converted WebDataset shards (Step 5 of the plan):
1. Decode a sample of each tier through both `webdataset` and HF `datasets` loaders.
2. Build the FLAME-params in-memory dict and confirm it joins to every view-tier scene via
ENV = EXPR_ENV[5:] (not a prefix match).
3. Confirm scene-mesh key is a true prefix of the corresponding view-tier keys.
4. Test itertools.groupby scene (30-view) and subject assembly, single-process/no-shuffle.
"""
import glob
import io
import itertools
import sys
import numpy as np
import trimesh
import webdataset as wds
from datasets import load_dataset
SHARD_DIR = sys.argv[1] if len(sys.argv) > 1 else "./data/shards_single_subject"
def view_key_parts(key):
subject, expr_env, cam = key.split("/")
expr, env = expr_env.split("_")
return subject, expr, env, cam
def check_decode_both_loaders():
print("=" * 60)
print("1. Decode check (webdataset lib + HF datasets) per tier")
print("=" * 60)
view_shards = sorted(glob.glob(f"{SHARD_DIR}/views/*.tar"))
mesh_shards = sorted(glob.glob(f"{SHARD_DIR}/scene_mesh/*.tar"))
flame_shards = sorted(glob.glob(f"{SHARD_DIR}/flame_params/*.tar"))
print(f"\n--- view tier: webdataset lib, first 3 samples ---")
ds = wds.WebDataset(view_shards, shardshuffle=False).decode()
for i, s in enumerate(ds):
rgb = np.array(s["rgb.png"]) if hasattr(s["rgb.png"], "shape") else None
print(f" key={s['__key__']} cam.json keys={list(s['cam.json'].keys())[:3]} "
f"landmark keys={list(s['landmark.npz'].keys())}")
if i >= 2:
break
print(f"\n--- view tier: HF datasets, first sample features ---")
hf_ds = load_dataset("webdataset", data_files=view_shards, split="train")
print(" features:", {k: str(v)[:60] for k, v in hf_ds.features.items()})
sample = hf_ds[0]
print(f" seg.png mode={sample['seg.png'].mode} size={sample['seg.png'].size}")
print(f" rgb.png mode={sample['rgb.png'].mode} size={sample['rgb.png'].size}")
assert sample["seg.png"].mode == "I;16", "segmentation did not decode as 16-bit!"
print(f"\n--- scene-mesh tier: webdataset lib ---")
ds = wds.WebDataset(mesh_shards, shardshuffle=False).decode()
for i, s in enumerate(ds):
mesh = trimesh.load(io.BytesIO(s["mesh.obj"] if isinstance(s["mesh.obj"], bytes) else s["mesh.obj"]), file_type="obj")
print(f" key={s['__key__']} mesh vertices={mesh.vertices.shape}")
if i >= 1:
break
print(f"\n--- flame-params tier: HF datasets ---")
hf_flame = load_dataset("webdataset", data_files=flame_shards, split="train")
print(" count:", len(hf_flame))
for row in hf_flame:
shape = np.array(row["flame.npz"]["expr"]).shape
print(f" key={row['__key__']} expr shape={shape}")
return view_shards, mesh_shards, flame_shards
def check_flame_join(view_shards, flame_shards):
print("\n" + "=" * 60)
print("2. FLAME-params join via ENV = EXPR_ENV[5:] (dict lookup, not prefix)")
print("=" * 60)
flame_ds = wds.WebDataset(flame_shards, shardshuffle=False).decode()
flame_by_key = {s["__key__"]: s["flame.npz"] for s in flame_ds}
print(f" loaded {len(flame_by_key)} flame-params entries: {list(flame_by_key.keys())}")
view_ds = wds.WebDataset(view_shards, shardshuffle=False)
scenes_seen = set()
missing = []
for s in view_ds:
subject, expr, env, cam = view_key_parts(s["__key__"])
expr_env = f"{expr}_{env}"
scene_key = f"{subject}/{expr_env}"
if scene_key in scenes_seen:
continue
scenes_seen.add(scene_key)
flame_key = f"{subject}/{env}"
if flame_key not in flame_by_key:
missing.append((scene_key, flame_key))
print(f" checked {len(scenes_seen)} distinct scenes")
if missing:
print(f" MISSING flame entries for: {missing}")
else:
print(" OK: every scene's FLAME-params entry found via env slice")
def check_scene_mesh_prefix(view_shards, mesh_shards):
print("\n" + "=" * 60)
print("3. Scene-mesh key is a true prefix of view-tier keys")
print("=" * 60)
mesh_keys = {s["__key__"] for s in wds.WebDataset(mesh_shards, shardshuffle=False)}
view_ds = wds.WebDataset(view_shards, shardshuffle=False)
bad = []
checked = 0
for s in view_ds:
subject, expr, env, cam = view_key_parts(s["__key__"])
scene_key = f"{subject}/{expr}_{env}"
checked += 1
if not s["__key__"].startswith(scene_key + "/"):
bad.append(s["__key__"])
if scene_key not in mesh_keys:
bad.append((s["__key__"], "no matching mesh key"))
print(f" checked {checked} view samples against {len(mesh_keys)} mesh keys")
print(" OK: prefix relationship holds" if not bad else f" FAILURES: {bad[:10]}")
def check_groupby(view_shards):
print("\n" + "=" * 60)
print("4. itertools.groupby scene (30-view) and subject assembly")
print(" (single-process, no shuffle, sequential read)")
print("=" * 60)
ds = wds.WebDataset(view_shards, shardshuffle=False)
def scene_key(sample):
subject, expr, env, cam = view_key_parts(sample["__key__"])
return f"{subject}/{expr}_{env}"
def subject_key(sample):
subject, expr, env, cam = view_key_parts(sample["__key__"])
return subject
scene_sizes = []
for key, group in itertools.groupby(ds, key=scene_key):
scene_sizes.append((key, len(list(group))))
print(f" found {len(scene_sizes)} scene groups")
bad_scenes = [(k, n) for k, n in scene_sizes if n != 30]
print(" OK: every scene has exactly 30 views" if not bad_scenes
else f" FAILURES (expected 30 views each): {bad_scenes}")
ds2 = wds.WebDataset(view_shards, shardshuffle=False)
subject_sizes = []
for key, group in itertools.groupby(ds2, key=subject_key):
subject_sizes.append((key, len(list(group))))
print(f" found {len(subject_sizes)} subject groups: {[(k, n) for k, n in subject_sizes]}")
if __name__ == "__main__":
view_shards, mesh_shards, flame_shards = check_decode_both_loaders()
check_flame_join(view_shards, flame_shards)
check_scene_mesh_prefix(view_shards, mesh_shards)
check_groupby(view_shards)
print("\nAll checks completed.")
|