dave.py

import json
import os

import datasets

_DESCRIPTION = "Dataset with video and audio references for epic and ego4d tasks."
_HOMEPAGE = "https://huggingface.co/datasets/gorjanradevski/dave"
_LICENSE = "MIT"

_MEDIA_FIELDS = [
    "compressed_video_path",
    "event_video_path",
    "video_with_overlayed_audio_path",
    "silent_video_path",
    "overlayed_audio_path",
]

def count_files_in_directory(directory):
    return sum(len(files) for _, _, files in os.walk(directory))

class DaveDataset(datasets.GeneratorBasedBuilder):
    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features({
                "compressed_video_path": datasets.Value("string"),
                "overlayed_event_index": datasets.Value("int32"),
                "events": [
                    {
                        "start": datasets.Value("string"),
                        "end": datasets.Value("string"),
                        "duration": datasets.Value("float64"),
                        "narration": datasets.Value("string"),
                        "action": datasets.Value("string"),
                        "raw_narration": datasets.Value("string"),
                    }
                ],
                "event_video_path": datasets.Value("string"),
                "audio_class": datasets.Value("string"),
                "video_with_overlayed_audio_path": datasets.Value("string"),
                "silent_video_path": datasets.Value("string"),
                "overlayed_audio_path": datasets.Value("string"),
                "video_id": datasets.Value("string"),
                "participant_id": datasets.Value("string"),
                "type": datasets.Value("string"),
                "raw_choices_simple_audio_classification": datasets.Sequence(datasets.Value("string")),
                "raw_choices_overlayed_full_audio_classification": datasets.Sequence(datasets.Value("string")),
                "raw_choices_video_segment": datasets.Sequence(datasets.Value("string")),
                "correct_temporal_order": datasets.Sequence(datasets.Value("string")),
                "raw_choices_temporal_video": datasets.Sequence(datasets.Value("string")),
                "raw_choices_multimodal": datasets.Sequence(datasets.Value("string")),
                "raw_choices_silent_video": datasets.Sequence(datasets.Value("string")),
                "raw_choices_audio": datasets.Sequence(datasets.Value("string")),
                "raw_choices_text_only": datasets.Sequence(datasets.Value("string")),
                "raw_choices_pipeline_event_classification": datasets.Sequence(datasets.Value("string")),
            }),
            homepage=_HOMEPAGE,
            license=_LICENSE,
        )

    def _split_generators(self, dl_manager):
        base_url = "https://huggingface.co/datasets/gorjanradevski/dave/resolve/main/"
        
        # Add both epic and ego4d files and json files
        json_files = {"ego4d": "ego4d.json", "epic": "epic.json"}
        zip_urls = {"ego4d": base_url + "ego4d.zip", "epic": base_url + "epic.zip"}
        
        split_generators = []
        for split_name, json_file in json_files.items():
            # Download JSON metadata file
            json_path = dl_manager.download(base_url + json_file)
            
            # Download and extract ZIP file
            print(f"Downloading and extracting {split_name}_files.zip...")
            extracted_dir = dl_manager.download_and_extract(zip_urls[split_name])

            print(f"Extracted to: {extracted_dir}")
            print(f"Total number of files extracted: {count_files_in_directory(extracted_dir)}")
            
            # Path to the extracted directory
            if isinstance(extracted_dir, str):
                files_dir = extracted_dir
            else:
                files_dir = extracted_dir[zip_urls[split_name]]
                
            split_generators.append(
                datasets.SplitGenerator(
                    name=split_name,
                    gen_kwargs={
                        "json_path": json_path,
                        "files_dir": files_dir,
                        "split_name": split_name,
                    },
                )
            )

        return split_generators

    def _generate_examples(self, json_path, files_dir, split_name):
        with open(json_path, "r", encoding="utf-8") as f:
            data = json.load(f)
        
        print(f"Processing {split_name} split with extracted files in {files_dir}")
        
        # Check for the respective files directory based on the split
        files_dir = os.path.join(files_dir, f"{split_name}_files")
        if not os.path.exists(files_dir):
            print(f"Warning: '{split_name}_files' directory not found in {files_dir}")
            print(f"Available directories: {os.listdir(files_dir)}")
            raise ValueError(f"Could not find '{split_name}_files' directory at {files_dir}")
        
        # Create a mapping of original file paths to local file paths
        file_mapping = {}
        
        for idx, item in enumerate(data):
            # Debug first item
            if idx == 0:
                print(f"Processing first item: {item['video_id'] if 'video_id' in item else 'unknown'}")
            
            # Replace file paths with local paths for all media fields
            all_fields_resolved = True
            for field in _MEDIA_FIELDS:
                if field not in item or not item[field]:
                    continue
                
                original_path = item[field]
                
                # Check if we already processed this file path
                if original_path in file_mapping:
                    item[field] = file_mapping[original_path]
                    continue
                
                # Extract file name and construct local path
                file_name = os.path.basename(original_path)
                local_path = os.path.join(files_dir, file_name)
                
                # Check if the file exists
                if os.path.exists(local_path):
                    item[field] = local_path
                    file_mapping[original_path] = local_path
                else:
                    print(f"Warning: File not found for {field}: {local_path}")
                    all_fields_resolved = False
                    break
            
            if all_fields_resolved:
                yield idx, item