File size: 7,721 Bytes
f0d6538 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 | """
Visualize 100 samples from new_policy7w_v2_reformat dataset.
Output: /mnt/bn/bohanzhainas1/jiashuo/viz/
- samples/sample_NNN/ -> video frames + info.json
- index.html -> browsable HTML overview
"""
import os, json, base64, glob
import pandas as pd
from pathlib import Path
DATA_DIR = "/mnt/hdfs/byte_tt_data_cu_vagcp/haogeng.liu/new_policy7w_v2_reformat"
OUT_DIR = Path("/mnt/bn/bohanzhainas1/jiashuo/viz")
N_SAMPLES = 100
PREVIEW_FRAMES = [0, 4, 8, 12, 15] # frames shown in HTML (out of 16)
# ── helpers ──────────────────────────────────────────────────────────────────
def save_frames(frames_b64: list, out_dir: Path, prefix: str) -> list[str]:
"""Save all frames to disk, return relative paths."""
paths = []
for i, b64 in enumerate(frames_b64):
raw = base64.b64decode(b64)
rel = f"{prefix}_frame_{i:02d}.jpg"
(out_dir / rel).write_bytes(raw)
paths.append(rel)
return paths
def parse_row(row) -> dict:
msgs = json.loads(row["messages"])
extra = json.loads(row["extra_info"])
user_content = msgs[0]["content"] # list of {type, video/text, ...}
asst_content = msgs[1]["content"] # list of {type, text}
# extract videos and texts from user turn
videos, user_texts = [], []
for item in user_content:
if item["type"] == "video":
videos.append(item["video"]) # list of 16 base64 strings
elif item["type"] == "text":
user_texts.append(item["text"])
elif item["type"] == "image":
videos.append([item["image"]]) # treat single-image as 1-frame video
assistant_text = " ".join(
c["text"] for c in asst_content if c.get("type") == "text"
)
return {
"videos": videos,
"user_texts": user_texts,
"assistant_text": assistant_text,
"extra_info": extra,
}
def build_sample_dir(idx: int, parsed: dict) -> dict:
"""Save frames + info.json, return metadata for HTML."""
sdir = OUT_DIR / "samples" / f"sample_{idx:03d}"
sdir.mkdir(parents=True, exist_ok=True)
frame_paths = {} # video_idx -> [relative paths]
for vi, frames in enumerate(parsed["videos"]):
paths = save_frames(frames, sdir, f"video{vi+1}")
frame_paths[vi] = paths
info = {
"user_texts": parsed["user_texts"],
"assistant": parsed["assistant_text"],
"extra_info": parsed["extra_info"],
"num_videos": len(parsed["videos"]),
"frames_per_video": [len(v) for v in parsed["videos"]],
}
(sdir / "info.json").write_text(json.dumps(info, ensure_ascii=False, indent=2))
return {"sdir": sdir, "frame_paths": frame_paths, "info": info}
# ── HTML generation ───────────────────────────────────────────────────────────
def img_tag(sdir: Path, rel_path: str, width=160) -> str:
abs_path = sdir / rel_path
b64 = base64.b64encode(abs_path.read_bytes()).decode()
return f'<img src="data:image/jpeg;base64,{b64}" width="{width}" style="border-radius:4px;margin:2px">'
def json_block(obj) -> str:
if isinstance(obj, str):
try:
obj = json.loads(obj)
except Exception:
pass
text = json.dumps(obj, ensure_ascii=False, indent=2) if not isinstance(obj, str) else obj
return f'<pre style="white-space:pre-wrap;word-break:break-word;font-size:12px;background:#f5f5f5;padding:8px;border-radius:4px;max-height:300px;overflow-y:auto">{text}</pre>'
def render_sample_html(idx: int, meta: dict) -> str:
sdir = meta["sdir"]
frame_paths = meta["frame_paths"]
info = meta["info"]
# video strips (only PREVIEW_FRAMES)
video_html = ""
for vi, paths in frame_paths.items():
preview = [paths[i] for i in PREVIEW_FRAMES if i < len(paths)]
imgs = "".join(img_tag(sdir, p) for p in preview)
video_html += f"""
<div style="margin-bottom:8px">
<b>Video {vi+1}</b> ({len(paths)} frames, showing {len(preview)}):
<div>{imgs}</div>
</div>"""
# user texts
user_text_html = ""
for i, t in enumerate(info["user_texts"]):
label = f"Video {i+1} metadata" if i < len(frame_paths) else f"Text {i+1}"
user_text_html += f"<b>{label}:</b>{json_block(t)}"
# assistant
asst_html = json_block(info["assistant_text"])
# extra_info
extra = info["extra_info"]
extra_sections = ""
for key in ["video1", "video2"]:
if key in extra:
v = extra[key]
extra_sections += f"<b>extra_info.{key}:</b>{json_block(v)}"
for key in [k for k in extra if k not in ("video1", "video2")]:
extra_sections += f"<b>extra_info.{key}:</b> <code>{extra[key]}</code><br>"
return f"""
<div id="s{idx}" style="border:1px solid #ddd;border-radius:8px;padding:16px;margin-bottom:24px;background:#fff">
<h3 style="margin:0 0 12px 0;color:#333">Sample {idx:03d}
<a href="samples/sample_{idx:03d}/info.json" target="_blank"
style="font-size:12px;margin-left:8px;color:#888">[info.json]</a>
</h3>
<div style="display:flex;gap:24px;flex-wrap:wrap">
<div style="flex:1;min-width:340px">
<b>Videos:</b>
{video_html}
</div>
<div style="flex:2;min-width:340px">
<b>User texts:</b>
{user_text_html}
<b>Assistant:</b>
{asst_html}
{extra_sections}
</div>
</div>
</div>"""
# ── main ─────────────────────────────────────────────────────────────────────
def main():
parquet_files = sorted(glob.glob(f"{DATA_DIR}/*.parquet"))
print(f"Found {len(parquet_files)} parquet files")
samples_html = []
count = 0
for pf in parquet_files:
if count >= N_SAMPLES:
break
df = pd.read_parquet(pf)
for _, row in df.iterrows():
if count >= N_SAMPLES:
break
print(f" Processing sample {count:03d} ...", end="\r")
try:
parsed = parse_row(row)
meta = build_sample_dir(count, parsed)
html = render_sample_html(count, meta)
samples_html.append(html)
except Exception as e:
samples_html.append(f'<div style="color:red">Sample {count}: ERROR - {e}</div>')
count += 1
print(f"\nProcessed {count} samples")
page = f"""<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<title>new_policy7w_v2_reformat — {count} samples</title>
<style>
body {{ font-family: sans-serif; max-width: 1400px; margin: 0 auto; padding: 24px; background: #f9f9f9; }}
h1 {{ color: #222; }}
code {{ background: #eee; padding: 2px 4px; border-radius: 3px; font-size: 12px; }}
</style>
</head>
<body>
<h1>Dataset: new_policy7w_v2_reformat</h1>
<p>Showing <b>{count}</b> samples |
Each sample: 2 videos (16 frames each) + metadata texts + assistant label<br>
Full frames + info.json saved under <code>samples/sample_NNN/</code></p>
{"".join(samples_html)}
</body>
</html>"""
out_html = OUT_DIR / "index.html"
out_html.write_text(page, encoding="utf-8")
print(f"Saved: {out_html}")
print(f"Samples dir: {OUT_DIR / 'samples'}")
if __name__ == "__main__":
main()
|