Spaces:
Running
on
Zero
Running
on
Zero
File size: 7,676 Bytes
a1d8bef |
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 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 |
import os
import json
import numpy as np
import webdataset as wds
import pytorch_lightning as pl
import torch
from torch.utils.data import Dataset
from torch.utils.data.distributed import DistributedSampler
from PIL import Image
from pathlib import Path
from tqdm import tqdm
from src.utils.train_util import instantiate_from_config
class DataModuleFromConfig(pl.LightningDataModule):
def __init__(
self,
batch_size=8,
num_workers=4,
train=None,
validation=None,
test=None,
**kwargs,
):
super().__init__()
self.batch_size = batch_size
self.num_workers = num_workers
self.dataset_configs = dict()
if train is not None:
self.dataset_configs['train'] = train
if validation is not None:
self.dataset_configs['validation'] = validation
if test is not None:
self.dataset_configs['test'] = test
def setup(self, stage):
if stage in ['fit']:
self.datasets = dict((k, instantiate_from_config(self.dataset_configs[k])) for k in self.dataset_configs)
else:
raise NotImplementedError
def train_dataloader(self):
# sampler = DistributedSampler(self.datasets['train'])
return wds.WebLoader(self.datasets['train'], batch_size=self.batch_size, num_workers=self.num_workers, shuffle=True)
def val_dataloader(self):
# sampler = DistributedSampler(self.datasets['validation'])
return wds.WebLoader(self.datasets['validation'], batch_size=4, num_workers=self.num_workers, shuffle=False)
def test_dataloader(self):
return wds.WebLoader(self.datasets['validation'], batch_size=self.batch_size, num_workers=self.num_workers, shuffle=False)
class RefinementData(Dataset):
lights_to_caption = {
0 : "Morning light",
1 : "Midday light, clear sky",
2 : "Afternoon light, cloudy sky",
}
def __init__(self,
root_dir='refinement_dataset/',
gt_subpath='gt',
pred_subpath='shap_e',
validation=False,
overfit=False,
caption_path=None,
split_path=None,
single_view=False,
single_light=False,
) -> None:
self.root_dir = Path(root_dir)
self.gt_subpath = gt_subpath
self.pred_subpath = pred_subpath
self.single_view = single_view
self.single_light = single_light
if caption_path is not None:
caption_path = self.root_dir / caption_path
with open(caption_path) as f:
self.captions_dict = json.load(f)
split_json = self.root_dir / split_path
with open(split_json) as f:
split_dict = json.load(f)
# print(split_dict.keys
# exit(0)
if validation:
uuids = split_dict['val']
else:
uuids = split_dict['train']
self.paths = [self.root_dir / uuid for uuid in uuids]
print('============= length of dataset %d =============' % len(self.paths))
def __len__(self):
return len(self.paths)
def load_im(self, path, color):
pil_img = Image.open(path)
image = np.asarray(pil_img, dtype=np.float32) / 255.
if image.shape[2] == 4:
alpha = image[:, :, 3:]
image = image[:, :, :3] * alpha + color * (1 - alpha)
else:
alpha = np.ones_like(image[:, :, :1])
image = torch.from_numpy(image).permute(2, 0, 1).contiguous().float()
alpha = torch.from_numpy(alpha).permute(2, 0, 1).contiguous().float()
return image, alpha
def __getitem__(self, index):
if os.path.exists(self.paths[index] / 'lights.json'):
num_lights = 3
else:
num_lights = len(os.listdir(self.paths[index] / self.gt_subpath))
if self.single_view:
view_index = np.random.randint(0, 6)
if self.single_light:
light_index = 0
else:
light_index = np.random.randint(0, num_lights)
# print("light index", light_index)
# exit(0)
uuid = self.paths[index].name
caption = self.captions_dict[uuid]
# if "lights.json" in os.listdir(self.paths[index]) and num_lights == 3: # according to additions to the dataset
# caption += " " + self.lights_to_caption[light_index]
image_path_gt = os.path.join(self.paths[index],'gt',str(light_index), "latent.pt")
image_path_pred = os.path.join(self.paths[index],'shap_e', "latent.pt")
'''background color, default: white'''
try:
imgs_gt = torch.load(image_path_gt,map_location='cpu').squeeze()
imgs_pred = torch.load(image_path_pred,map_location='cpu').squeeze()
except Exception as e:
print("Error loading latent tensors, gt path %s, pred path %s" % (image_path_gt, image_path_pred))
raise e
if self.single_view:
row = view_index // 2
col = view_index % 2
imgs_gt = imgs_gt[:, row*40:(row+1)*40, col*40:(col+1)*40]
imgs_pred = imgs_pred[:, row*40:(row+1)*40, col*40:(col+1)*40]
# imgs_gt = imgs_gt
data = {
'refined_imgs': imgs_gt, # (6, 3, H, W)
'unrefined_imgs': imgs_pred, # (6, 3, H, W)
'caption': caption,
'index': index
}
return data
class ObjaverseData(Dataset):
def __init__(self,
root_dir='objaverse/',
meta_fname='valid_paths.json',
image_dir='rendering_zero123plus',
validation=False,
):
self.root_dir = Path(root_dir)
self.image_dir = image_dir
with open(os.path.join(root_dir, meta_fname)) as f:
lvis_dict = json.load(f)
paths = []
for k in lvis_dict.keys():
paths.extend(lvis_dict[k])
self.paths = paths
total_objects = len(self.paths)
if validation:
self.paths = self.paths[-16:] # used last 16 as validation
else:
self.paths = self.paths[:-16]
print('============= length of dataset %d =============' % len(self.paths))
def __len__(self):
return len(self.paths)
def load_im(self, path, color):
pil_img = Image.open(path)
image = np.asarray(pil_img, dtype=np.float32) / 255.
alpha = image[:, :, 3:]
image = image[:, :, :3] * alpha + color * (1 - alpha)
image = torch.from_numpy(image).permute(2, 0, 1).contiguous().float()
alpha = torch.from_numpy(alpha).permute(2, 0, 1).contiguous().float()
return image, alpha
def __getitem__(self, index):
while True:
image_path = os.path.join(self.root_dir, self.image_dir, self.paths[index])
'''background color, default: white'''
bkg_color = [1., 1., 1.]
img_list = []
try:
for idx in range(7):
img, alpha = self.load_im(os.path.join(image_path, '%03d.png' % idx), bkg_color)
img_list.append(img)
except Exception as e:
print(e)
index = np.random.randint(0, len(self.paths))
continue
break
imgs = torch.stack(img_list, dim=0).float()
data = {
'cond_imgs': imgs[0], # (3, H, W)
'target_imgs': imgs[1:], # (6, 3, H, W)
}
return data
|