Spaces:
Runtime error
Runtime error
import os | |
import os.path | |
import random | |
import numpy as np | |
import torch | |
import torch.utils.data as data | |
from PIL import Image | |
class DeepFashionAttrPoseDataset(data.Dataset): | |
def __init__(self, | |
pose_dir, | |
texture_ann_dir, | |
shape_ann_path, | |
downsample_factor=2, | |
xflip=False): | |
self._densepose_path = pose_dir | |
self._image_fnames_target = [] | |
self._image_fnames = [] | |
self.upper_fused_attrs = [] | |
self.lower_fused_attrs = [] | |
self.outer_fused_attrs = [] | |
self.shape_attrs = [] | |
self.downsample_factor = downsample_factor | |
self.xflip = xflip | |
# load attributes | |
assert os.path.exists(f'{texture_ann_dir}/upper_fused.txt') | |
for idx, row in enumerate( | |
open(os.path.join(f'{texture_ann_dir}/upper_fused.txt'), 'r')): | |
annotations = row.split() | |
self._image_fnames_target.append(annotations[0]) | |
self._image_fnames.append(f'{annotations[0].split(".")[0]}.png') | |
self.upper_fused_attrs.append(int(annotations[1])) | |
assert len(self._image_fnames_target) == len(self.upper_fused_attrs) | |
assert os.path.exists(f'{texture_ann_dir}/lower_fused.txt') | |
for idx, row in enumerate( | |
open(os.path.join(f'{texture_ann_dir}/lower_fused.txt'), 'r')): | |
annotations = row.split() | |
assert self._image_fnames_target[idx] == annotations[0] | |
self.lower_fused_attrs.append(int(annotations[1])) | |
assert len(self._image_fnames_target) == len(self.lower_fused_attrs) | |
assert os.path.exists(f'{texture_ann_dir}/outer_fused.txt') | |
for idx, row in enumerate( | |
open(os.path.join(f'{texture_ann_dir}/outer_fused.txt'), 'r')): | |
annotations = row.split() | |
assert self._image_fnames_target[idx] == annotations[0] | |
self.outer_fused_attrs.append(int(annotations[1])) | |
assert len(self._image_fnames_target) == len(self.outer_fused_attrs) | |
assert os.path.exists(shape_ann_path) | |
for idx, row in enumerate(open(os.path.join(shape_ann_path), 'r')): | |
annotations = row.split() | |
assert self._image_fnames_target[idx] == annotations[0] | |
self.shape_attrs.append([int(i) for i in annotations[1:]]) | |
def _open_file(self, path_prefix, fname): | |
return open(os.path.join(path_prefix, fname), 'rb') | |
def _load_densepose(self, raw_idx): | |
fname = self._image_fnames[raw_idx] | |
fname = f'{fname[:-4]}_densepose.png' | |
with self._open_file(self._densepose_path, fname) as f: | |
densepose = Image.open(f) | |
if self.downsample_factor != 1: | |
width, height = densepose.size | |
width = width // self.downsample_factor | |
height = height // self.downsample_factor | |
densepose = densepose.resize( | |
size=(width, height), resample=Image.NEAREST) | |
# channel-wise IUV order, [3, H, W] | |
densepose = np.array(densepose)[:, :, 2:].transpose(2, 0, 1) | |
return densepose.astype(np.float32) | |
def __getitem__(self, index): | |
pose = self._load_densepose(index) | |
shape_attr = self.shape_attrs[index] | |
shape_attr = torch.LongTensor(shape_attr) | |
if self.xflip and random.random() > 0.5: | |
pose = pose[:, :, ::-1].copy() | |
upper_fused_attr = self.upper_fused_attrs[index] | |
lower_fused_attr = self.lower_fused_attrs[index] | |
outer_fused_attr = self.outer_fused_attrs[index] | |
pose = pose / 12. - 1 | |
return_dict = { | |
'densepose': pose, | |
'img_name': self._image_fnames_target[index], | |
'shape_attr': shape_attr, | |
'upper_fused_attr': upper_fused_attr, | |
'lower_fused_attr': lower_fused_attr, | |
'outer_fused_attr': outer_fused_attr, | |
} | |
return return_dict | |
def __len__(self): | |
return len(self._image_fnames) | |