File size: 19,750 Bytes
65600d1
 
 
 
 
 
 
 
 
 
 
9d63714
65600d1
 
 
 
 
b21e8c5
65600d1
d42fccb
 
 
 
 
 
 
 
 
 
25c33f6
 
 
 
 
 
 
 
 
 
 
65600d1
b21e8c5
 
25c33f6
65600d1
b21e8c5
65600d1
b21e8c5
65600d1
 
 
 
 
b21e8c5
 
 
 
 
 
65600d1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b21e8c5
 
65600d1
 
d42fccb
 
 
 
 
 
 
 
 
 
 
 
 
 
349e3bf
d42fccb
 
 
 
349e3bf
 
65600d1
25c33f6
 
65600d1
 
 
9d63714
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d42fccb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
86d12ed
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d42fccb
65600d1
a94b543
d42fccb
a94b543
65600d1
d42fccb
 
 
 
 
 
 
 
 
 
65600d1
 
 
 
a94b543
65600d1
 
86d12ed
 
 
65600d1
86d12ed
65600d1
 
a94b543
 
 
 
 
 
 
 
 
 
 
 
 
65600d1
 
6bd8e43
65600d1
9461a66
 
65600d1
 
 
 
 
 
6ed909f
 
 
 
 
 
9461a66
 
 
65600d1
9461a66
 
 
 
 
6bd8e43
9461a66
6bd8e43
9461a66
 
 
 
 
86d12ed
9461a66
6bd8e43
 
86d12ed
9461a66
 
 
 
 
 
6bd8e43
86d12ed
 
d42fccb
86d12ed
 
 
d42fccb
86d12ed
d42fccb
 
 
 
 
 
 
 
 
 
 
 
 
86d12ed
 
 
 
0bdde06
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
86d12ed
9d63714
 
 
 
 
 
 
86d12ed
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c8edbbe
 
 
 
 
 
 
 
 
 
 
86d12ed
 
 
 
 
 
 
 
 
 
afef5d4
70110f0
afef5d4
 
70110f0
 
afef5d4
 
70110f0
afef5d4
70110f0
 
afef5d4
 
70110f0
 
afef5d4
 
 
 
86d12ed
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c8edbbe
 
afef5d4
 
 
 
c8edbbe
 
 
 
 
 
 
 
afef5d4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c8edbbe
 
 
86d12ed
 
 
 
 
 
 
 
 
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
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
"""
Load data from Hugging Face dataset AE-W/batch_outputs.
Uses huggingface_hub to list and download files on demand.
"""
import json
import os
import re
from pathlib import Path
from typing import Optional

from huggingface_hub import HfApi, hf_hub_download, list_repo_files
from huggingface_hub import HfFileSystem


REPO_ID = "AE-W/batch_outputs"
REPO_TYPE = "dataset"
ROOT_PREFIX = "batch_outputs/"
DASHENG_PREFIX = "batch_outputs_dasheng/"

# Three methods: each has batch_outputs_* + generated_noises_*
BIN_BATCH_PREFIX = "batch_outputs_bin/"
BIN_GENERATED_PREFIX = "generated_noises_bin/"
CLAP_BATCH_PREFIX = "batch_outputs/"  # ROOT_PREFIX
CLAP_GENERATED_PREFIX = "generated_noises_clap/"
DASHENG_BATCH_PREFIX = "batch_outputs_dasheng/"
DASHENG_GENERATED_PREFIX = "generated_noises_dasheng/"

GENERATED_SKIP_IDS = {"__pycache__", "NearestNeighbor_space_push"}

# Cache full repo file list so we only call list_repo_files once per process (major speedup)
_cached_repo_files: Optional[list[str]] = None


def _get_repo_files() -> list[str]:
    """Return full list of repo file paths, cached after first call."""
    global _cached_repo_files
    if _cached_repo_files is None:
        _cached_repo_files = list_repo_files(REPO_ID, repo_type=REPO_TYPE)
    return _cached_repo_files


def _get_sample_ids(prefix: str = ROOT_PREFIX) -> list[str]:
    """List sample IDs (e.g. 07_003277) under given prefix in repo."""
    files = _get_repo_files()
    seen = set()
    pat = re.escape(prefix.rstrip("/")) + r"/([^/]+)/"
    for f in files:
        m = re.match(pat, f)
        if m:
            seen.add(m.group(1))
    return sorted(seen)


def _get_all_sample_ids() -> list[str]:
    """Union of sample IDs from batch_outputs and batch_outputs_dasheng."""
    ids = set(_get_sample_ids(ROOT_PREFIX)) | set(_get_sample_ids(DASHENG_PREFIX))
    return sorted(ids)


def _download_file(path_in_repo: str, local_dir: Optional[str] = None) -> str:
    """Download a file from the dataset; return local path."""
    return hf_hub_download(
        repo_id=REPO_ID,
        filename=path_in_repo,
        repo_type=REPO_TYPE,
        local_dir=local_dir,
        force_download=False,
    )


def _load_json_from_repo(path_in_repo: str) -> Optional[list]:
    """Download and load JSON file from repo."""
    try:
        path = _download_file(path_in_repo)
        with open(path, "r", encoding="utf-8") as f:
            return json.load(f)
    except Exception:
        return None


def list_samples() -> list[str]:
    """Return list of sample IDs (bid) from both batch_outputs and batch_outputs_dasheng."""
    return _get_all_sample_ids()


def list_samples_bin() -> list[str]:
    """Sample IDs for Bin: batch_outputs_bin ∪ generated_noises_bin (exclude __pycache__ etc)."""
    batch_ids = set(_get_sample_ids(BIN_BATCH_PREFIX))
    gen_ids = set(_get_sample_ids(BIN_GENERATED_PREFIX)) - GENERATED_SKIP_IDS
    return sorted(batch_ids | gen_ids)


def list_samples_clap() -> list[str]:
    """Sample IDs for Clap: batch_outputs ∪ generated_noises_clap."""
    batch_ids = set(_get_sample_ids(CLAP_BATCH_PREFIX))
    gen_ids = set(_get_sample_ids(CLAP_GENERATED_PREFIX)) - GENERATED_SKIP_IDS
    return sorted(batch_ids | gen_ids)


def list_samples_dasheng() -> list[str]:
    """Sample IDs for Dasheng: batch_outputs_dasheng (excl. fold*) ∪ generated_noises_dasheng."""
    batch_ids = {x for x in _get_sample_ids(DASHENG_BATCH_PREFIX) if not x.startswith("fold")}
    gen_ids = set(_get_sample_ids(DASHENG_GENERATED_PREFIX)) - GENERATED_SKIP_IDS
    return sorted(batch_ids | gen_ids)


def _find_files(inner: str) -> list[str]:
    """List all repo files under inner path (uses cached repo file list)."""
    files = _get_repo_files()
    return [f for f in files if f.startswith(inner + "/")]


def _list_files_under_via_fs(path_in_repo: str) -> list[str]:
    """List files under path_in_repo using HfFileSystem (avoids relying on full list_repo_files)."""
    try:
        fs = HfFileSystem()
        fs_prefix = f"datasets/{REPO_ID}/{path_in_repo}"
        strip = f"datasets/{REPO_ID}/"
        found = fs.glob(fs_prefix + "/**")
        out = []
        for p in found:
            if p.endswith("/"):
                continue
            rel = p.replace(strip, "").lstrip("/")
            if rel.startswith(path_in_repo + "/"):
                out.append(rel)
        return out
    except Exception:
        return []


def _has_files_under(prefix_bid: str) -> bool:
    """True if repo has any file under prefix_bid/."""
    files = _get_repo_files()
    return any(f.startswith(prefix_bid + "/") for f in files)


def resolve_sample_prefix(bid: str, method: str) -> Optional[str]:
    """
    Resolve which repo prefix contains this sample_id (batch first, then generated).
    method: "bin" | "clap" | "dasheng"
    Returns e.g. "batch_outputs_bin/" or "generated_noises_bin/".
    """
    if method == "bin":
        batch, generated = BIN_BATCH_PREFIX, BIN_GENERATED_PREFIX
    elif method == "clap":
        batch, generated = CLAP_BATCH_PREFIX, CLAP_GENERATED_PREFIX
    elif method == "dasheng":
        batch, generated = DASHENG_BATCH_PREFIX, DASHENG_GENERATED_PREFIX
    else:
        return None
    if _has_files_under(batch + bid):
        return batch
    if _has_files_under(generated + bid):
        return generated
    return None


def get_inner_path(prefix: str, bid: str) -> Optional[str]:
    """
    Return the inner path (contains baseline/, natural_prompts.json, etc.).
    For batch_outputs*: prefix/bid/bid. For generated_noises*: same or prefix/bid/X/X if bid/X/X.
    """
    inner_std = f"{prefix}{bid}/{bid}"
    if _has_files_under(inner_std):
        return inner_std
    # generated_noises: may have prefix/bid/X/X (e.g. cars_honking/heavy_machinery/heavy_machinery)
    if not prefix.startswith("generated_noises"):
        return inner_std
    files = _get_repo_files()
    for f in files:
        if not f.startswith(prefix + bid + "/"):
            continue
        if "natural_prompts.json" in f or "temp_retrieval.json" in f:
            parts = f.split("/")
            # prefix/bid/X/X/file -> inner = prefix/bid/X/X
            if len(parts) >= 4:
                return "/".join(parts[:4])
    return inner_std


def _collect_block(file_list: list, folder_prefix: str) -> dict:
    """From files under folder_prefix, get spec + bg_wav, fg_wav, m_wav."""
    spec = bg = fg = m = None
    for f in file_list:
        if folder_prefix not in f:
            continue
        name = f.split("/")[-1]
        if name.endswith(".png"):
            spec = f
        elif name.endswith("_bg.wav"):
            bg = f
        elif name.endswith("_fg.wav"):
            fg = f
        elif name.endswith("_m.wav"):
            m = f
    return {
        "spec": _download_file(spec) if spec else None,
        "bg_wav": _download_file(bg) if bg else None,
        "fg_wav": _download_file(fg) if fg else None,
        "m_wav": _download_file(m) if m else None,
    }


def get_nn_demo_paths(bid: str, top_k: int = 10, root_prefix: Optional[str] = None, method: Optional[str] = None) -> dict:
    """
    For NN view: NN1-NN10 from baseline (generated_baseline_01, 02, ..., 10) in prompt order.
    root_prefix: legacy; if method is set (bin|clap|dasheng), resolve prefix and inner from repo.
    Returns {nn_list: [{spec, bg_wav, fg_wav, m_wav, prompt, similarity}, ...]}.
    """
    if method is not None:
        prefix = resolve_sample_prefix(bid, method)
        if not prefix:
            return {"nn_list": []}
        inner = get_inner_path(prefix, bid)
        if not inner:
            return {"nn_list": []}
    else:
        prefix = root_prefix if root_prefix is not None else ROOT_PREFIX
        inner = f"{prefix}{bid}/{bid}"
    prompts = _load_json_from_repo(f"{inner}/temp_retrieval.json")
    if not prompts:
        prompts = _load_json_from_repo(f"{inner}/natural_prompts.json")
    if not prompts:
        return {"nn_list": []}

    files = _find_files(inner)
    baseline_inner = f"{inner}/baseline"
    baseline_files = _find_files(baseline_inner) if any(f.startswith(baseline_inner) for f in files) else []

    nn_list = []
    for i, p in enumerate(prompts[:top_k]):
        prompt = p.get("prompt", "")
        sim = p.get("similarity_score", p.get("retrieval_score"))
        bl_prefix = f"generated_baseline_{i+1:02d}_"
        block = {"spec": None, "bg_wav": None, "fg_wav": None, "m_wav": None}
        for f in baseline_files:
            parts = f.replace(baseline_inner + "/", "").split("/")
            if parts and parts[0].startswith(bl_prefix):
                full_prefix = baseline_inner + "/" + parts[0]
                block = _collect_block(baseline_files, full_prefix)
                break
        block["prompt"] = prompt
        block["similarity"] = sim
        nn_list.append(block)

    return {"nn_list": nn_list}


def get_noise_demo_paths(bid: str) -> dict:
    """
    One block per prompt (1, 2, 3): each has prompt text, baseline (spec + 3 wavs), and our method (spec + 3 wavs).
    Returns { "block1": {prompt, baseline: {...}, nn: {...}}, "block2": ..., "block3": ... }.
    """
    inner = f"{ROOT_PREFIX}{bid}/{bid}"
    files = _find_files(inner)
    baseline_inner = f"{inner}/baseline"
    baseline_files = _find_files(baseline_inner) if any(f.startswith(baseline_inner) for f in files) else []

    prompts = _load_json_from_repo(f"{inner}/temp_retrieval.json")
    if not prompts:
        prompts = _load_json_from_repo(f"{inner}/natural_prompts.json")
    if not prompts:
        prompts = []

    # Find baseline folder names generated_baseline_01_*, 02_*, 03_*
    seen = set()
    baseline_folders = []
    for f in baseline_files:
        parts = f.replace(baseline_inner + "/", "").split("/")
        if parts and parts[0].startswith("generated_baseline_") and parts[0] not in seen:
            seen.add(parts[0])
            baseline_folders.append((parts[0], baseline_inner + "/" + parts[0]))
    baseline_folders.sort(key=lambda x: x[0])

    result = {}
    for i in range(1, 4):
        prompt_text = prompts[i - 1].get("prompt", "") if i <= len(prompts) else ""
        bl_prefix = f"generated_baseline_{i:02d}_"
        baseline_block = {"spec": None, "bg_wav": None, "fg_wav": None, "m_wav": None}
        for folder_name, full_prefix in baseline_folders:
            if folder_name.startswith(bl_prefix):
                baseline_block = _collect_block(baseline_files, full_prefix)
                break
        rel_prefix = f"generated_{i:02d}_"
        nn_files = [f for f in files if f.replace(inner + "/", "").startswith(rel_prefix)]
        nn_block = _collect_block(nn_files, rel_prefix)
        nn_block["prompt"] = prompt_text
        result[f"block{i}"] = {
            "prompt": prompt_text,
            "baseline": baseline_block,
            "nn": nn_block,
        }
    return result


def get_results_demo_paths(bid: str, root_prefix: Optional[str] = None, method: Optional[str] = None) -> dict:
    """
    For Results view: 3 blocks (prompts 1-3), each with 4 columns:
    Baseline (original), Gaussian, Youtube-noise, Ours.
    root_prefix: legacy; if method is set (bin|clap|dasheng), resolve prefix and inner from repo.
    """
    if method is not None:
        prefix = resolve_sample_prefix(bid, method)
        if not prefix:
            return {}
        inner = get_inner_path(prefix, bid)
        if not inner:
            return {}
        # Dasheng-style: prompt-named folders (batch_outputs_bin, batch_outputs_dasheng, all generated_noises_*)
        use_dasheng = prefix in (BIN_BATCH_PREFIX, DASHENG_BATCH_PREFIX, DASHENG_GENERATED_PREFIX) or prefix.startswith("generated_noises")
    else:
        prefix = root_prefix if root_prefix is not None else ROOT_PREFIX
        inner = f"{prefix}{bid}/{bid}"
        use_dasheng = root_prefix == DASHENG_PREFIX
    files = _find_files(inner)
    baseline_inner = f"{inner}/baseline"
    gaussian_inner = f"{inner}/gaussian_baseline"
    youtube_inner = f"{inner}/youtube_noise_baseline"
    # Use full repo file list for baseline/gaussian/youtube so we find them even if "files" is partial
    all_repo = _get_repo_files()
    baseline_files = _find_files(baseline_inner) if any(f.startswith(baseline_inner + "/") for f in all_repo) else []
    gaussian_files = _find_files(gaussian_inner) if any(f.startswith(gaussian_inner + "/") for f in all_repo) else []
    youtube_files = _find_files(youtube_inner) if any(f.startswith(youtube_inner + "/") for f in all_repo) else []

    # Fallback for bin/generated: gaussian or youtube may live under a different inner (e.g. prefix/bid/X/X)
    if not gaussian_files and (prefix == BIN_BATCH_PREFIX or prefix.startswith("generated_noises")):
        for f in all_repo:
            if f.startswith(prefix + bid + "/") and "/gaussian_baseline/" in f and (f.endswith("_m.wav") or f.endswith("_bg.wav")):
                gaussian_inner_fb = f.rsplit("/", 1)[0]  # path to gaussian_baseline dir
                gaussian_files = _find_files(gaussian_inner_fb)
                gaussian_inner = gaussian_inner_fb
                break
    if not youtube_files and (prefix == BIN_BATCH_PREFIX or prefix.startswith("generated_noises")):
        for f in all_repo:
            if f.startswith(prefix + bid + "/") and "/youtube_noise_baseline/" in f and (f.endswith("_m.wav") or f.endswith("_bg.wav")):
                # f = .../youtube_noise_baseline/<prompt_folder>/file_m.wav -> parent = youtube_noise_baseline
                youtube_inner_fb = f.split("/youtube_noise_baseline/", 1)[0] + "/youtube_noise_baseline"
                youtube_files = _find_files(youtube_inner_fb)
                youtube_inner = youtube_inner_fb
                break

    # For generated_noises: list_repo_files() may not include these paths in large repos; use HfFileSystem by path
    if prefix.startswith("generated_noises"):
        if not gaussian_files:
            gaussian_files = _list_files_under_via_fs(gaussian_inner)
        if not youtube_files:
            youtube_files = _list_files_under_via_fs(youtube_inner)

    prompts = _load_json_from_repo(f"{inner}/temp_retrieval.json")
    if not prompts:
        prompts = _load_json_from_repo(f"{inner}/natural_prompts.json")
    if not prompts:
        prompts = []

    def get_baseline_folders(bl_inner, bl_files):
        seen = set()
        folders = []
        for f in bl_files:
            parts = f.replace(bl_inner + "/", "").split("/")
            if parts and parts[0].startswith("generated_baseline_") and parts[0] not in seen:
                seen.add(parts[0])
                folders.append((parts[0], bl_inner + "/" + parts[0]))
        folders.sort(key=lambda x: x[0])
        return folders

    def get_youtube_folders():
        if use_dasheng:
            # Dasheng: subdirs are prompt names (underscores)
            seen = set()
            folders = []
            for f in youtube_files:
                parts = f.replace(youtube_inner + "/", "").split("/")
                if parts and parts[0] not in seen:
                    seen.add(parts[0])
                    folders.append((parts[0], youtube_inner + "/" + parts[0]))
            folders.sort(key=lambda x: x[0])
            return folders
        seen = set()
        folders = []
        for f in youtube_files:
            parts = f.replace(youtube_inner + "/", "").split("/")
            if parts and parts[0].startswith("generated_") and parts[0] not in seen:
                seen.add(parts[0])
                folders.append((parts[0], youtube_inner + "/" + parts[0]))
        folders.sort(key=lambda x: x[0])
        return folders

    def _match_dasheng_folder(folder_name: str, folders: list[tuple[str, str]]) -> Optional[tuple[str, str]]:
        """Match prompt-derived folder_name to actual folder; allow truncated names and hyphen/underscore."""
        if not folder_name or not folders:
            return None
        # Normalize: prompt may have "ground-level" / "low-intensity" but dir is "ground_level" / "low_inte"
        normalized = folder_name.replace("-", "_")
        # Exact match
        for fn, fp in folders:
            if fn == folder_name or fn == normalized:
                return (fn, fp)
        # Folder may be truncated: actual fn is prefix of folder_name (e.g. fn="..._low_inte", folder_name="..._low_intensity_...")
        candidates = [(fn, fp) for fn, fp in folders if normalized.startswith(fn) or folder_name.startswith(fn)]
        if candidates:
            return max(candidates, key=lambda x: len(x[0]))
        # Or folder_name (or normalized) is prefix of fn
        candidates = [(fn, fp) for fn, fp in folders if fn.startswith(normalized) or fn.startswith(folder_name)]
        if candidates:
            return min(candidates, key=lambda x: len(x[0]))
        return None

    baseline_folders = get_baseline_folders(baseline_inner, baseline_files)
    youtube_folders = get_youtube_folders()

    result = {}
    for i in range(1, 4):
        prompt_text = prompts[i - 1].get("prompt", "") if i <= len(prompts) else ""
        bl_prefix = f"generated_baseline_{i:02d}_"
        rel_prefix = f"generated_{i:02d}_"

        bl_orig = {"spec": None, "bg_wav": None, "fg_wav": None, "m_wav": None}
        for fn, fp in baseline_folders:
            if fn.startswith(bl_prefix):
                bl_orig = _collect_block(baseline_files, fp)
                break

        gaussian_block = _collect_block(gaussian_files, gaussian_inner)

        bl_youtube = {"spec": None, "bg_wav": None, "fg_wav": None, "m_wav": None}
        if use_dasheng:
            folder_name = prompt_text.replace(" ", "_") if prompt_text else ""
            matched = _match_dasheng_folder(folder_name, youtube_folders)
            if matched:
                fn, fp = matched
                bl_youtube = _collect_block(youtube_files, fp)
        else:
            for fn, fp in youtube_folders:
                if fn.startswith(rel_prefix):
                    bl_youtube = _collect_block(youtube_files, fp)
                    break

        if use_dasheng:
            folder_name = prompt_text.replace(" ", "_") if prompt_text else ""
            # Ours: list prompt-named dirs under inner (exclude baseline, gaussian_baseline, youtube_noise_baseline)
            skip = {"baseline", "youtube_noise_baseline", "gaussian_baseline"}
            inner_dirs = set()
            for f in files:
                if not f.startswith(inner + "/"):
                    continue
                rest = f.replace(inner + "/", "", 1)
                if "/" in rest:
                    top = rest.split("/")[0]
                    if top not in skip and not top.startswith("generated_baseline"):
                        inner_dirs.add(top)
            inner_folders = [(d, inner + "/" + d) for d in sorted(inner_dirs)]
            ours_fn_fp = _match_dasheng_folder(folder_name, inner_folders)
            if ours_fn_fp:
                fn, fp = ours_fn_fp
                nn_files = [f for f in files if f.startswith(fp + "/")]
                ours_block = _collect_block(nn_files, fp)
            else:
                ours_block = {"spec": None, "bg_wav": None, "fg_wav": None, "m_wav": None}
        else:
            nn_files = [f for f in files if f.replace(inner + "/", "").startswith(rel_prefix)]
            ours_block = _collect_block(nn_files, inner + "/" + rel_prefix)

        result[f"block{i}"] = {
            "prompt": prompt_text,
            "baseline_original": bl_orig,
            "baseline_gaussian": gaussian_block,
            "baseline_youtube": bl_youtube,
            "ours": ours_block,
        }
    return result