File size: 1,622 Bytes
032e687 |
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 |
import os
from mmengine.dist import master_only
from vlm.datasets.evaluation.base_eval_dataset import BaseEvalDataset
class LLaVAPretrainDataset(BaseEvalDataset):
METAINFO: dict = dict(name='llava_pretrain')
def __init__(
self,
image_folder,
split=8,
rank=0,
):
super().__init__()
self.image_files = os.listdir(image_folder)
# self.image_paths = [os.path.join(image_folder, file_name) for file_name in self.image_files]
self.image_paths = self._get_image_paths(image_folder)
assert rank < split
size = len(self.image_paths) // split + 1
i_s, i_e = size * rank, min(size * (rank + 1), len(self.image_paths))
self.image_paths = self.image_paths[i_s: i_e]
def _get_image_paths(self, folder_path):
ret = []
sub_files = os.listdir(folder_path)
for file_name in sub_files:
if '.png' not in file_name and '.jpg' not in file_name:
# a folder
sub_folder_files = self._get_image_paths(os.path.join(folder_path, file_name))
ret.extend(sub_folder_files)
else:
ret.append(os.path.join(folder_path, file_name))
return ret
def __len__(self):
return len(self.image_paths)
def __getitem__(self, idx):
data_dict = {'image_path': self.image_paths[idx]}
data_dict.update({'text_prompts': self.image_paths[idx], 'pixel_values': None, 'img_id': str(idx)})
return data_dict
@master_only
def evaluate(self, **kwargs):
return {'Acc': 0}
|