|
|
--- |
|
|
dataset_info: |
|
|
features: |
|
|
- name: model_name |
|
|
dtype: string |
|
|
- name: id_prompt |
|
|
dtype: string |
|
|
- name: frame_prompt |
|
|
dtype: string |
|
|
- name: Image |
|
|
dtype: image |
|
|
- name: sub_images |
|
|
list: |
|
|
- name: bytes |
|
|
dtype: binary |
|
|
- name: videos |
|
|
list: |
|
|
- name: motion_bucket_id |
|
|
dtype: int64 |
|
|
- name: video_bytes |
|
|
dtype: binary |
|
|
splits: |
|
|
- name: train |
|
|
num_bytes: 45316656.0 |
|
|
num_examples: 2 |
|
|
download_size: 45321507 |
|
|
dataset_size: 45316656.0 |
|
|
configs: |
|
|
- config_name: default |
|
|
data_files: |
|
|
- split: train |
|
|
path: data/train-* |
|
|
--- |
|
|
|
|
|
```python |
|
|
from datasets import Dataset, load_dataset |
|
|
from PIL import Image |
|
|
import io |
|
|
|
|
|
def bytes_to_image(image_bytes): |
|
|
""" |
|
|
将字节数据(bytes)转换为 PIL.Image 对象。 |
|
|
""" |
|
|
image_stream = io.BytesIO(image_bytes) |
|
|
image = Image.open(image_stream) |
|
|
return image |
|
|
|
|
|
# 加载数据集 |
|
|
ds = load_dataset("svjack/OnePromptOneStory-Examples-Vid-head2")["train"] |
|
|
|
|
|
# 定义 motion_bucket_ids |
|
|
motion_bucket_ids = [10, 20, 30, 40, 50] |
|
|
|
|
|
# 创建一个新的数据集列表 |
|
|
new_data = [] |
|
|
|
|
|
# 遍历原始数据集中的每一行 |
|
|
for example in ds: |
|
|
sub_images = example["sub_images"] |
|
|
videos = example["videos"] |
|
|
|
|
|
# 遍历每个 sub_image |
|
|
for idx, sub_image_dict in enumerate(sub_images): |
|
|
sub_image_bytes = sub_image_dict["bytes"] |
|
|
sub_image = bytes_to_image(sub_image_bytes) |
|
|
|
|
|
# 计算对应的视频索引 |
|
|
video_idx = idx * len(motion_bucket_ids) |
|
|
|
|
|
# 遍历每个 motion_bucket_id 和对应的视频 |
|
|
for i, motion_bucket_id in enumerate(motion_bucket_ids): |
|
|
video_dict = videos[video_idx + i] |
|
|
video_bytes = video_dict["video_bytes"] # 视频的二进制数据 |
|
|
|
|
|
# 创建新的样本,保留原始数据的所有字段 |
|
|
new_sample = { |
|
|
**example, # 保留原始数据的所有字段 |
|
|
"sub_image": sub_image, |
|
|
"motion_bucket_id": motion_bucket_id, |
|
|
"video": video_bytes # 直接存储视频的二进制数据 |
|
|
} |
|
|
|
|
|
# 添加到新数据集中 |
|
|
new_data.append(new_sample) |
|
|
|
|
|
# 将新数据转换为 Hugging Face Dataset 对象 |
|
|
new_dataset = Dataset.from_list(new_data) |
|
|
|
|
|
# 查看新数据集中的第一个样本 |
|
|
#print(new_dataset[0]) |
|
|
|
|
|
new_dataset.push_to_hub("svjack/OnePromptOneStory-Examples-Vid-head2-Exp") |
|
|
``` |
|
|
|