| import os |
| import json |
| import numpy as np |
| import pandas as pd |
| from tqdm import tqdm |
| import librosa |
|
|
| |
| stride = 20 |
| motion_length = 64 |
| speaker_target = 2 |
| use_additional = False |
|
|
| root_dir = './beat_english_v2.0.0/' |
| output_dir = "./datasets/data_json/" |
| os.makedirs(output_dir, exist_ok=True) |
| train_test_split_path = './beat_english_v2.0.0/train_test_split.csv' |
| df = pd.read_csv(train_test_split_path) |
|
|
| filtered_df = df[(df['id'].str.split('_').str[0].astype(int) == speaker_target) & (df['type'] != 'additional')] |
| clips = [] |
| for idx, row_item in tqdm(filtered_df.iterrows()): |
| video_id = row_item['id'] |
| mode = row_item['type'] |
| |
| npz_path = os.path.join(root_dir, "smplxflame_30", video_id + ".npz") |
| wav_path = os.path.join(root_dir, "wave16k", video_id + ".wav") |
|
|
| try: |
| motion_data = np.load(npz_path, allow_pickle=True) |
| except: |
| print(f"cant open {npz_path}") |
| |
| try: |
| wave_data, _ = librosa.load(wav_path, sr=None) |
| except: |
| print(f"cant open {wav_path}") |
|
|
| motion = motion_data['poses'] |
| total_len = motion.shape[0] |
|
|
| for i in range(0, total_len - motion_length, stride): |
| clip = { |
| "video_id": video_id, |
| "motion_path": npz_path, |
| "audio_path": wav_path, |
| "mode": mode, |
| "start_idx": i, |
| "end_idx": i + motion_length |
| } |
| clips.append(clip) |
|
|
| output_json = os.path.join(output_dir, f"beat2_s{stride}_l{motion_length}_speaker{speaker_target}.json") |
| with open(output_json, 'w') as f: |
| json.dump(clips, f, indent=4) |
|
|
