| |
| """Example: Validate schema and timeline sanity for one row.""" |
| from datasets import load_dataset |
|
|
| REPO = "skyzhou06/StreamingOmniDatasets" |
| REQUIRED_FIELDS = [ |
| "id", "dataset", "task_family", "source_dataset", "source_id", |
| "split", "duration_sec", "modalities", "timeline", "evidence", |
| "quality", "metadata", |
| ] |
| TIMELINE_FIELDS = [ |
| "t_start", "t_end", "audio_ref", "frame_refs", "asr_delta", |
| "ocr_delta", "audio_events", "visual_events", "control", |
| "target_text_delta", "target_speech_ref", |
| ] |
|
|
| def validate_row(sample, config_name): |
| issues = [] |
|
|
| |
| missing = [f for f in REQUIRED_FIELDS if f not in sample] |
| if missing: |
| issues.append(f"Missing schema fields: {missing}") |
|
|
| |
| timeline = sample.get("timeline", []) |
| if not timeline: |
| issues.append("timeline is empty") |
| else: |
| t0 = timeline[0] |
| missing_tl = [f for f in TIMELINE_FIELDS if f not in t0] |
| if missing_tl: |
| issues.append(f"Missing timeline fields: {missing_tl}") |
|
|
| |
| if config_name == "AVJointReasoning": |
| if "question" not in sample: |
| issues.append("Missing AVJointReasoning: question") |
| if "answer" not in sample: |
| issues.append("Missing AVJointReasoning: answer") |
|
|
| |
| for i, chunk in enumerate(timeline): |
| t_start = chunk.get("t_start", 0) |
| t_end = chunk.get("t_end", 0) |
| if t_start > t_end: |
| issues.append(f"Timeline chunk {i}: t_start > t_end") |
|
|
| return issues |
|
|
| for config in ["StreamingAudioInteraction", "AVJointReasoning"]: |
| print(f"\nValidating {config}...") |
| try: |
| ds = load_dataset(REPO, config, split="train", streaming=True) |
| sample = next(iter(ds)) |
| issues = validate_row(sample, config) |
| if issues: |
| for issue in issues: |
| print(f" ISSUE: {issue}") |
| else: |
| print(f" All checks passed for {sample.get('id')}") |
| except Exception as e: |
| print(f" Loading failed: {e}") |
| print(" Try parquet fallback as shown in load_streaming_omni_dataset.py") |
|
|
