mattdeitke
commited on
Commit
•
3dc86cc
1
Parent(s):
aae8281
convert objaverse 1.0 to use the abstract class
Browse files- objaverse_xl/objaverse_v1.py +431 -285
objaverse_xl/objaverse_v1.py
CHANGED
@@ -10,100 +10,151 @@ import requests
|
|
10 |
import pandas as pd
|
11 |
import tempfile
|
12 |
from objaverse_xl.utils import get_file_hash
|
|
|
13 |
|
14 |
import fsspec
|
15 |
from loguru import logger
|
16 |
from tqdm import tqdm
|
17 |
|
18 |
|
19 |
-
|
20 |
-
"""
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
fs.
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
44 |
|
45 |
-
|
46 |
-
|
47 |
-
|
|
|
|
|
48 |
|
49 |
-
|
|
|
|
|
50 |
|
51 |
-
|
|
|
52 |
|
|
|
53 |
|
54 |
-
def
|
55 |
-
|
56 |
-
download_dir: str = "~/.objaverse",
|
57 |
-
) -> Dict[str, Any]:
|
58 |
-
"""Load the full metadata of all objects in the dataset.
|
59 |
|
60 |
-
|
61 |
-
|
62 |
-
the metadata for all uids.
|
63 |
-
download_dir: The base directory to download the annotations to. Supports all
|
64 |
-
file systems supported by fsspec. Defaults to "~/.objaverse".
|
65 |
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
metadata_path = os.path.join(download_dir, "hf-objaverse-v1", "metadata")
|
72 |
-
fs, _ = fsspec.core.url_to_fs(metadata_path)
|
73 |
-
fs.makedirs(metadata_path, exist_ok=True)
|
74 |
-
|
75 |
-
# get the dir ids that need to be loaded if only downloading a subset of uids
|
76 |
-
object_paths = _load_object_paths(download_dir=download_dir)
|
77 |
-
dir_ids = (
|
78 |
-
{object_paths[uid].split("/")[1] for uid in uids}
|
79 |
-
if uids is not None
|
80 |
-
else {f"{i // 1000:03d}-{i % 1000:03d}" for i in range(160)}
|
81 |
-
)
|
82 |
-
|
83 |
-
# get the existing metadata files
|
84 |
-
existing_metadata_files = fs.glob(
|
85 |
-
os.path.join(metadata_path, "*.json.gz"), refresh=True
|
86 |
-
)
|
87 |
-
existing_dir_ids = {
|
88 |
-
file.split("/")[-1].split(".")[0]
|
89 |
-
for file in existing_metadata_files
|
90 |
-
if file.endswith(".json.gz") # note partial files end with .json.gz.tmp
|
91 |
-
}
|
92 |
-
downloaded_dir_ids = existing_dir_ids.intersection(dir_ids)
|
93 |
-
logger.info(f"Found {len(downloaded_dir_ids)} metadata files already downloaded")
|
94 |
-
|
95 |
-
# download the metadata from the missing dir_ids
|
96 |
-
dir_ids_to_download = dir_ids - existing_dir_ids
|
97 |
-
logger.info(f"Downloading {len(dir_ids_to_download)} metadata files")
|
98 |
-
|
99 |
-
# download the metadata file if it doesn't exist
|
100 |
-
if len(dir_ids_to_download) > 0:
|
101 |
-
for i_id in tqdm(dir_ids_to_download, desc="Downloading metadata files"):
|
102 |
-
# get the path to the json file
|
103 |
-
path = os.path.join(metadata_path, f"{i_id}.json.gz")
|
104 |
|
105 |
-
|
106 |
-
|
|
|
|
|
|
|
107 |
|
108 |
# download the file to a tmp path to avoid partial downloads on interruption
|
109 |
tmp_path = f"{path}.tmp"
|
@@ -112,235 +163,330 @@ def load_full_annotations(
|
|
112 |
f.write(response.read())
|
113 |
fs.rename(tmp_path, path)
|
114 |
|
115 |
-
|
116 |
-
for i_id in tqdm(dir_ids, desc="Reading metadata files"):
|
117 |
-
# get the path to the json file
|
118 |
-
path = os.path.join(metadata_path, f"{i_id}.json.gz")
|
119 |
-
|
120 |
-
# read the json file of the metadata chunk
|
121 |
with fs.open(path, "rb") as f:
|
122 |
with gzip.GzipFile(fileobj=f) as gfile:
|
123 |
content = gfile.read()
|
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 |
-
uid: The uid of the object to load.
|
191 |
-
hf_object_path: The path to the object in the Hugging Face repo. Here, hf_object_path
|
192 |
-
is the part that comes after "main" in the Hugging Face repo url:
|
193 |
-
https://huggingface.co/datasets/allenai/objaverse/resolve/main/{hf_object_path}
|
194 |
-
download_dir: The base directory to download the object to. Supports all
|
195 |
-
file systems supported by fsspec. Defaults to "~/.objaverse".
|
196 |
|
197 |
-
|
198 |
-
|
199 |
-
|
200 |
-
hf_url = f"https://huggingface.co/datasets/allenai/objaverse/resolve/main/{hf_object_path}"
|
201 |
-
|
202 |
-
filename = os.path.join(download_dir, "hf-objaverse-v1", hf_object_path)
|
203 |
-
fs, path = fsspec.core.url_to_fs(filename)
|
204 |
-
|
205 |
-
# download the file
|
206 |
-
fs.makedirs(os.path.dirname(path), exist_ok=True)
|
207 |
-
tmp_path = f"{path}.tmp"
|
208 |
-
with fs.open(tmp_path, "wb") as file:
|
209 |
-
with urllib.request.urlopen(hf_url) as response:
|
210 |
-
file.write(response.read())
|
211 |
-
|
212 |
-
fs.rename(tmp_path, path)
|
213 |
-
|
214 |
-
return uid, filename
|
215 |
-
|
216 |
-
|
217 |
-
def _parallel_download_object(args):
|
218 |
-
# workaround since starmap doesn't work well with tqdm
|
219 |
-
return _download_object(*args)
|
220 |
-
|
221 |
-
|
222 |
-
def load_objects(
|
223 |
-
uids: List[str],
|
224 |
-
download_processes: int = 1,
|
225 |
-
download_dir: str = "~/.objaverse",
|
226 |
-
) -> Dict[str, str]:
|
227 |
-
"""Return the path to the object files for the given uids.
|
228 |
-
|
229 |
-
If the object is not already downloaded, it will be downloaded.
|
230 |
-
|
231 |
-
Args:
|
232 |
-
uids: A list of uids.
|
233 |
-
download_processes: The number of processes to use to download the objects.
|
234 |
-
|
235 |
-
Returns:
|
236 |
-
A dictionary mapping the object uid to the local path of where the object
|
237 |
-
downloaded.
|
238 |
-
"""
|
239 |
-
uids_set = set(uids)
|
240 |
-
hf_object_paths = _load_object_paths(download_dir=download_dir)
|
241 |
-
|
242 |
-
versioned_dirname = os.path.join(download_dir, "hf-objaverse-v1")
|
243 |
-
fs, path = fsspec.core.url_to_fs(versioned_dirname)
|
244 |
-
|
245 |
-
# Get the existing file paths. This is much faster than calling fs.exists() for each
|
246 |
-
# file. `glob()` is like walk, but returns a list of files instead of the nested
|
247 |
-
# directory structure. glob() is also faster than find() / walk() since it doesn't
|
248 |
-
# need to traverse the entire directory structure.
|
249 |
-
existing_file_paths = fs.glob(
|
250 |
-
os.path.join(path, "glbs", "*", "*.glb"), refresh=True
|
251 |
-
)
|
252 |
-
existing_uids = {
|
253 |
-
file.split("/")[-1].split(".")[0]
|
254 |
-
for file in existing_file_paths
|
255 |
-
if file.endswith(".glb") # note partial files end with .glb.tmp
|
256 |
-
}
|
257 |
-
|
258 |
-
# add the existing downloaded uids to the return dict
|
259 |
-
out = {}
|
260 |
-
already_downloaded_uids = uids_set.intersection(existing_uids)
|
261 |
-
for uid in already_downloaded_uids:
|
262 |
-
hf_object_path = hf_object_paths[uid]
|
263 |
-
fs_abs_object_path = os.path.join(versioned_dirname, hf_object_path)
|
264 |
-
out[uid] = fs_abs_object_path
|
265 |
-
|
266 |
-
logger.info(f"Found {len(already_downloaded_uids)} objects already downloaded")
|
267 |
-
|
268 |
-
# get the uids that need to be downloaded
|
269 |
-
remaining_uids = uids_set - existing_uids
|
270 |
-
uids_to_download = []
|
271 |
-
for uid in remaining_uids:
|
272 |
-
if uid not in hf_object_paths:
|
273 |
-
logger.error(f"Could not find object with uid {uid}. Skipping it.")
|
274 |
-
continue
|
275 |
-
uids_to_download.append((uid, hf_object_paths[uid]))
|
276 |
-
|
277 |
-
logger.info(f"Downloading {len(uids_to_download)} new objects")
|
278 |
-
|
279 |
-
# check if all objects are already downloaded
|
280 |
-
if len(uids_to_download) == 0:
|
281 |
-
return out
|
282 |
|
283 |
-
if download_processes == 1:
|
284 |
-
# iteratively download the objects
|
285 |
-
for uid, hf_object_path in tqdm(uids_to_download):
|
286 |
-
uid, local_path = _download_object(
|
287 |
-
uid=uid, hf_object_path=hf_object_path, download_dir=download_dir
|
288 |
-
)
|
289 |
-
out[uid] = local_path
|
290 |
-
else:
|
291 |
args = [
|
292 |
-
(
|
293 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
294 |
]
|
295 |
|
296 |
# download the objects in parallel
|
297 |
-
with Pool(
|
298 |
new_object_downloads = list(
|
299 |
tqdm(
|
300 |
-
pool.imap_unordered(_parallel_download_object, args),
|
301 |
total=len(args),
|
302 |
)
|
303 |
)
|
304 |
|
305 |
-
for
|
306 |
-
out[
|
307 |
-
|
308 |
-
return out
|
309 |
|
|
|
310 |
|
311 |
-
def load_lvis_annotations(
|
312 |
-
|
|
|
|
|
|
|
313 |
|
314 |
-
|
315 |
|
316 |
-
|
317 |
-
|
318 |
-
|
319 |
|
320 |
-
|
321 |
-
|
322 |
-
|
323 |
-
|
324 |
|
325 |
-
|
326 |
-
|
327 |
-
|
328 |
|
329 |
-
|
330 |
-
|
331 |
-
|
332 |
-
|
333 |
-
|
334 |
|
335 |
-
|
336 |
-
|
337 |
-
|
338 |
-
|
339 |
|
340 |
-
|
341 |
-
|
342 |
-
|
343 |
-
|
344 |
-
|
345 |
|
346 |
-
|
|
|
10 |
import pandas as pd
|
11 |
import tempfile
|
12 |
from objaverse_xl.utils import get_file_hash
|
13 |
+
from objaverse_xl.abstract import ObjaverseSource
|
14 |
|
15 |
import fsspec
|
16 |
from loguru import logger
|
17 |
from tqdm import tqdm
|
18 |
|
19 |
|
20 |
+
class Sketchfab(ObjaverseSource):
|
21 |
+
"""A class for downloading and processing Objaverse 1.0."""
|
22 |
+
|
23 |
+
def load_annotations(self, download_dir: str = "~/.objaverse") -> pd.DataFrame:
|
24 |
+
"""Load the annotations from the given directory.
|
25 |
+
|
26 |
+
Args:
|
27 |
+
download_dir (str, optional): The directory to load the annotations from.
|
28 |
+
Supports all file systems supported by fsspec. Defaults to
|
29 |
+
"~/.objaverse".
|
30 |
+
|
31 |
+
Returns:
|
32 |
+
pd.DataFrame: The annotations, which includes the columns "thingId", "fileId",
|
33 |
+
"filename", and "license".
|
34 |
+
"""
|
35 |
+
remote_url = "https://huggingface.co/datasets/allenai/objaverse-xl/resolve/main/objaverse_v1/object-metadata.parquet"
|
36 |
+
download_path = os.path.join(
|
37 |
+
download_dir, "hf-objaverse-v1", "thingiverse-objects.parquet"
|
38 |
+
)
|
39 |
+
fs, path = fsspec.core.url_to_fs(download_path)
|
40 |
+
|
41 |
+
if not fs.exists(path):
|
42 |
+
fs.makedirs(os.path.dirname(path), exist_ok=True)
|
43 |
+
logger.info(f"Downloading {remote_url} to {download_path}")
|
44 |
+
response = requests.get(remote_url)
|
45 |
+
response.raise_for_status()
|
46 |
+
with fs.open(path, "wb") as file:
|
47 |
+
file.write(response.content)
|
48 |
+
|
49 |
+
# read the file with pandas and fsspec
|
50 |
+
with fs.open(download_path, "rb") as f:
|
51 |
+
annotations_df = pd.read_parquet(f)
|
52 |
+
|
53 |
+
annotations_df["metadata"] = "{}"
|
54 |
+
|
55 |
+
return annotations_df
|
56 |
+
|
57 |
+
def load_full_annotations(
|
58 |
+
self,
|
59 |
+
uids: Optional[List[str]] = None,
|
60 |
+
download_dir: str = "~/.objaverse",
|
61 |
+
) -> Dict[str, Any]:
|
62 |
+
"""Load the full metadata of all objects in the dataset.
|
63 |
+
|
64 |
+
Args:
|
65 |
+
uids: A list of uids with which to load metadata. If None, it loads
|
66 |
+
the metadata for all uids.
|
67 |
+
download_dir: The base directory to download the annotations to. Supports all
|
68 |
+
file systems supported by fsspec. Defaults to "~/.objaverse".
|
69 |
+
|
70 |
+
Returns:
|
71 |
+
A dictionary of the metadata for each object. The keys are the uids and the
|
72 |
+
values are the metadata for that object.
|
73 |
+
"""
|
74 |
+
# make the metadata dir if it doesn't exist
|
75 |
+
metadata_path = os.path.join(download_dir, "hf-objaverse-v1", "metadata")
|
76 |
+
fs, _ = fsspec.core.url_to_fs(metadata_path)
|
77 |
+
fs.makedirs(metadata_path, exist_ok=True)
|
78 |
+
|
79 |
+
# get the dir ids that need to be loaded if only downloading a subset of uids
|
80 |
+
object_paths = self._load_object_paths(download_dir=download_dir)
|
81 |
+
dir_ids = (
|
82 |
+
{object_paths[uid].split("/")[1] for uid in uids}
|
83 |
+
if uids is not None
|
84 |
+
else {f"{i // 1000:03d}-{i % 1000:03d}" for i in range(160)}
|
85 |
+
)
|
86 |
+
|
87 |
+
# get the existing metadata files
|
88 |
+
existing_metadata_files = fs.glob(
|
89 |
+
os.path.join(metadata_path, "*.json.gz"), refresh=True
|
90 |
+
)
|
91 |
+
existing_dir_ids = {
|
92 |
+
file.split("/")[-1].split(".")[0]
|
93 |
+
for file in existing_metadata_files
|
94 |
+
if file.endswith(".json.gz") # note partial files end with .json.gz.tmp
|
95 |
+
}
|
96 |
+
downloaded_dir_ids = existing_dir_ids.intersection(dir_ids)
|
97 |
+
logger.info(
|
98 |
+
f"Found {len(downloaded_dir_ids)} metadata files already downloaded"
|
99 |
+
)
|
100 |
+
|
101 |
+
# download the metadata from the missing dir_ids
|
102 |
+
dir_ids_to_download = dir_ids - existing_dir_ids
|
103 |
+
logger.info(f"Downloading {len(dir_ids_to_download)} metadata files")
|
104 |
+
|
105 |
+
# download the metadata file if it doesn't exist
|
106 |
+
if len(dir_ids_to_download) > 0:
|
107 |
+
for i_id in tqdm(dir_ids_to_download, desc="Downloading metadata files"):
|
108 |
+
# get the path to the json file
|
109 |
+
path = os.path.join(metadata_path, f"{i_id}.json.gz")
|
110 |
+
|
111 |
+
# get the url to the remote json file
|
112 |
+
hf_url = f"https://huggingface.co/datasets/allenai/objaverse/resolve/main/metadata/{i_id}.json.gz"
|
113 |
+
|
114 |
+
# download the file to a tmp path to avoid partial downloads on interruption
|
115 |
+
tmp_path = f"{path}.tmp"
|
116 |
+
with fs.open(tmp_path, "wb") as f:
|
117 |
+
with urllib.request.urlopen(hf_url) as response:
|
118 |
+
f.write(response.read())
|
119 |
+
fs.rename(tmp_path, path)
|
120 |
+
|
121 |
+
out = {}
|
122 |
+
for i_id in tqdm(dir_ids, desc="Reading metadata files"):
|
123 |
+
# get the path to the json file
|
124 |
+
path = os.path.join(metadata_path, f"{i_id}.json.gz")
|
125 |
|
126 |
+
# read the json file of the metadata chunk
|
127 |
+
with fs.open(path, "rb") as f:
|
128 |
+
with gzip.GzipFile(fileobj=f) as gfile:
|
129 |
+
content = gfile.read()
|
130 |
+
data = json.loads(content)
|
131 |
|
132 |
+
# filter the data to only include the uids we want
|
133 |
+
if uids is not None:
|
134 |
+
data = {uid: data[uid] for uid in uids if uid in data}
|
135 |
|
136 |
+
# add the data to the out dict
|
137 |
+
out.update(data)
|
138 |
|
139 |
+
return out
|
140 |
|
141 |
+
def _load_object_paths(self, download_dir: str) -> Dict[str, str]:
|
142 |
+
"""Load the object paths from the dataset.
|
|
|
|
|
|
|
143 |
|
144 |
+
The object paths specify the location of where the object is located in the
|
145 |
+
Hugging Face repo.
|
|
|
|
|
|
|
146 |
|
147 |
+
Returns:
|
148 |
+
A dictionary mapping the uid to the object path.
|
149 |
+
"""
|
150 |
+
object_paths_file = "object-paths.json.gz"
|
151 |
+
local_path = os.path.join(download_dir, "hf-objaverse-v1", object_paths_file)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
152 |
|
153 |
+
# download the object_paths file if it doesn't exist
|
154 |
+
fs, path = fsspec.core.url_to_fs(local_path)
|
155 |
+
if not fs.exists(path):
|
156 |
+
hf_url = f"https://huggingface.co/datasets/allenai/objaverse/resolve/main/{object_paths_file}"
|
157 |
+
fs.makedirs(os.path.dirname(path), exist_ok=True)
|
158 |
|
159 |
# download the file to a tmp path to avoid partial downloads on interruption
|
160 |
tmp_path = f"{path}.tmp"
|
|
|
163 |
f.write(response.read())
|
164 |
fs.rename(tmp_path, path)
|
165 |
|
166 |
+
# read the object_paths
|
|
|
|
|
|
|
|
|
|
|
167 |
with fs.open(path, "rb") as f:
|
168 |
with gzip.GzipFile(fileobj=f) as gfile:
|
169 |
content = gfile.read()
|
170 |
+
object_paths = json.loads(content)
|
171 |
+
|
172 |
+
return object_paths
|
173 |
+
|
174 |
+
def load_uids(self, download_dir: str = "~/.objaverse") -> List[str]:
|
175 |
+
"""Load the uids from the dataset.
|
176 |
+
|
177 |
+
Returns:
|
178 |
+
A list of all the UIDs from the dataset.
|
179 |
+
"""
|
180 |
+
return list(self._load_object_paths(download_dir=download_dir).keys())
|
181 |
+
|
182 |
+
def _download_object(
|
183 |
+
self,
|
184 |
+
file_identifier: str,
|
185 |
+
hf_object_path: str,
|
186 |
+
download_dir: Optional[str],
|
187 |
+
expected_sha256: str,
|
188 |
+
handle_found_object: Optional[Callable] = None,
|
189 |
+
handle_modified_object: Optional[Callable] = None,
|
190 |
+
handle_missing_object: Optional[Callable] = None,
|
191 |
+
) -> Tuple[str, Optional[str]]:
|
192 |
+
"""Download the object for the given uid.
|
193 |
+
|
194 |
+
Args:
|
195 |
+
file_identifier: The file identifier of the object.
|
196 |
+
hf_object_path: The path to the object in the Hugging Face repo. Here,
|
197 |
+
hf_object_path is the part that comes after "main" in the Hugging Face
|
198 |
+
repo url:
|
199 |
+
https://huggingface.co/datasets/allenai/objaverse/resolve/main/{hf_object_path}
|
200 |
+
download_dir: The base directory to download the object to. Supports all
|
201 |
+
file systems supported by fsspec. Defaults to "~/.objaverse".
|
202 |
+
expected_objects (str): The expected SHA256 of the contents of the
|
203 |
+
downloaded object.
|
204 |
+
handle_found_object (Optional[Callable], optional): Called when an object is
|
205 |
+
successfully found and downloaded. Here, the object has the same sha256
|
206 |
+
as the one that was downloaded with Objaverse-XL. If None, the object
|
207 |
+
will be downloaded, but nothing will be done with it. Args for the
|
208 |
+
function include:
|
209 |
+
- local_path (str): Local path to the downloaded 3D object.
|
210 |
+
- file_identifier (str): GitHub URL of the 3D object.
|
211 |
+
- sha256 (str): SHA256 of the contents of the 3D object.
|
212 |
+
- metadata (Dict[str, Any]): Metadata about the 3D object, including the
|
213 |
+
GitHub organization and repo names.
|
214 |
+
Return is not used. Defaults to None.
|
215 |
+
handle_modified_object (Optional[Callable], optional): Called when a
|
216 |
+
modified object is found and downloaded. Here, the object is
|
217 |
+
successfully downloaded, but it has a different sha256 than the one that
|
218 |
+
was downloaded with Objaverse-XL. This is not expected to happen very
|
219 |
+
often, because the same commit hash is used for each repo. If None, the
|
220 |
+
object will be downloaded, but nothing will be done with it. Args for
|
221 |
+
the function include:
|
222 |
+
- local_path (str): Local path to the downloaded 3D object.
|
223 |
+
- file_identifier (str): GitHub URL of the 3D object.
|
224 |
+
- new_sha256 (str): SHA256 of the contents of the newly downloaded 3D
|
225 |
+
object.
|
226 |
+
- old_sha256 (str): Expected SHA256 of the contents of the 3D object as
|
227 |
+
it was when it was downloaded with Objaverse-XL.
|
228 |
+
- metadata (Dict[str, Any]): Metadata about the 3D object, including the
|
229 |
+
GitHub organization and repo names.
|
230 |
+
Return is not used. Defaults to None.
|
231 |
+
handle_missing_object (Optional[Callable], optional): Called when an object
|
232 |
+
that is in Objaverse-XL is not found. Here, it is likely that the
|
233 |
+
repository was deleted or renamed. If None, nothing will be done with
|
234 |
+
the missing object. Args for the function include:
|
235 |
+
- file_identifier (str): GitHub URL of the 3D object.
|
236 |
+
- sha256 (str): SHA256 of the contents of the original 3D object.
|
237 |
+
- metadata (Dict[str, Any]): Metadata about the 3D object, including the
|
238 |
+
GitHub organization and repo names.
|
239 |
+
Return is not used. Defaults to None.
|
240 |
+
|
241 |
+
Returns:
|
242 |
+
A tuple of the uid and the path to where the downloaded object. If
|
243 |
+
download_dir is None, the path will be None.
|
244 |
+
"""
|
245 |
+
hf_url = f"https://huggingface.co/datasets/allenai/objaverse/resolve/main/{hf_object_path}"
|
246 |
+
|
247 |
+
with tempfile.TemporaryDirectory() as temp_dir:
|
248 |
+
# download the file locally
|
249 |
+
temp_path = os.path.join(temp_dir, hf_object_path)
|
250 |
+
os.makedirs(os.path.dirname(temp_path), exist_ok=True)
|
251 |
+
temp_path_tmp = f"{temp_path}.tmp"
|
252 |
+
with open(temp_path_tmp, "wb") as file:
|
253 |
+
with urllib.request.urlopen(hf_url) as response:
|
254 |
+
file.write(response.read())
|
255 |
+
os.rename(temp_path_tmp, temp_path)
|
256 |
+
|
257 |
+
# get the sha256 of the downloaded file
|
258 |
+
sha256 = get_file_hash(temp_path)
|
259 |
+
|
260 |
+
if sha256 == expected_sha256:
|
261 |
+
if handle_found_object is not None:
|
262 |
+
handle_found_object(
|
263 |
+
local_path=temp_path,
|
264 |
+
file_identifier=file_identifier,
|
265 |
+
sha256=sha256,
|
266 |
+
metadata=dict(),
|
267 |
+
)
|
268 |
+
else:
|
269 |
+
if handle_modified_object is not None:
|
270 |
+
handle_modified_object(
|
271 |
+
local_path=temp_path,
|
272 |
+
file_identifier=file_identifier,
|
273 |
+
new_sha256=sha256,
|
274 |
+
old_sha256=expected_sha256,
|
275 |
+
metadata=dict(),
|
276 |
+
)
|
277 |
+
|
278 |
+
if download_dir is not None:
|
279 |
+
filename = os.path.join(download_dir, "hf-objaverse-v1", hf_object_path)
|
280 |
+
fs, path = fsspec.core.url_to_fs(filename)
|
281 |
+
fs.makedirs(os.path.dirname(path), exist_ok=True)
|
282 |
+
fs.put(temp_path, path)
|
283 |
+
else:
|
284 |
+
path = None
|
285 |
+
|
286 |
+
return file_identifier, path
|
287 |
+
|
288 |
+
def _parallel_download_object(self, args):
|
289 |
+
# workaround since starmap doesn't work well with tqdm
|
290 |
+
return self._download_object(*args)
|
291 |
+
|
292 |
+
def _get_uid(self, item: pd.Series) -> str:
|
293 |
+
file_identifier = item["fileIdentifier"]
|
294 |
+
return file_identifier.split("/")[-1]
|
295 |
+
|
296 |
+
def uid_to_file_identifier(self, uid: str) -> str:
|
297 |
+
"""Convert the uid to the file identifier.
|
298 |
+
|
299 |
+
Args:
|
300 |
+
uid (str): The uid of the object.
|
301 |
+
|
302 |
+
Returns:
|
303 |
+
The file identifier of the object.
|
304 |
+
"""
|
305 |
+
return f"https://sketchfab.com/3d-models/{uid}"
|
306 |
+
|
307 |
+
def file_identifier_to_uid(self, file_identifier: str) -> str:
|
308 |
+
"""Convert the file identifier to the uid.
|
309 |
+
|
310 |
+
Args:
|
311 |
+
file_identifier (str): The file identifier of the object.
|
312 |
+
|
313 |
+
Returns:
|
314 |
+
The uid of the object.
|
315 |
+
"""
|
316 |
+
return file_identifier.split("/")[-1]
|
317 |
+
|
318 |
+
def download_objects(
|
319 |
+
self,
|
320 |
+
objects: pd.DataFrame,
|
321 |
+
download_dir: Optional[str] = "~/.objaverse",
|
322 |
+
processes: Optional[int] = None,
|
323 |
+
handle_found_object: Optional[Callable] = None,
|
324 |
+
handle_modified_object: Optional[Callable] = None,
|
325 |
+
handle_missing_object: Optional[Callable] = None,
|
326 |
+
**kwargs,
|
327 |
+
) -> Dict[str, str]:
|
328 |
+
"""Return the path to the object files for the given uids.
|
329 |
+
|
330 |
+
If the object is not already downloaded, it will be downloaded.
|
331 |
+
|
332 |
+
Args:
|
333 |
+
objects (pd.DataFrame): Objects to download. Must have columns for
|
334 |
+
the object "fileIdentifier" and "sha256". Use the `load_annotations`
|
335 |
+
function to get the metadata.
|
336 |
+
processes (int, optional): The number of processes to use to download
|
337 |
+
the objects. Defaults to 1.
|
338 |
+
download_dir (Optional[str], optional): The base directory to download the
|
339 |
+
object to. Supports all file systems supported by fsspec. If None, the
|
340 |
+
objects will be removed after downloading. Defaults to "~/.objaverse".
|
341 |
+
|
342 |
+
Returns:
|
343 |
+
A dictionary mapping the object fileIdentifier to the local path of where
|
344 |
+
the object downloaded.
|
345 |
+
"""
|
346 |
+
hf_object_paths = self._load_object_paths(
|
347 |
+
download_dir=download_dir if download_dir is not None else "~/.objaverse"
|
348 |
+
)
|
349 |
+
|
350 |
+
# make a copy of the objects so we don't modify the original
|
351 |
+
objects = objects.copy()
|
352 |
+
objects["uid"] = objects.apply(self._get_uid, axis=1)
|
353 |
+
uids_to_sha256 = dict(zip(objects["uid"], objects["sha256"]))
|
354 |
+
uids_set = set(uids_to_sha256.keys())
|
355 |
+
|
356 |
+
# create a new df where the uids are the index
|
357 |
+
objects_uid_index = objects.set_index("uid")
|
358 |
+
|
359 |
+
out = {}
|
360 |
+
objects_to_download = []
|
361 |
+
if download_dir is None:
|
362 |
+
for _, item in objects.iterrows():
|
363 |
+
uid = item["uid"]
|
364 |
+
if uid not in hf_object_paths:
|
365 |
+
logger.error(f"Could not find object with uid {uid}!")
|
366 |
+
if handle_missing_object is not None:
|
367 |
+
handle_missing_object(
|
368 |
+
file_identifier=item["fileIdentifier"],
|
369 |
+
sha256=item["sha256"],
|
370 |
+
metadata=dict(),
|
371 |
+
)
|
372 |
+
continue
|
373 |
+
objects_to_download.append(
|
374 |
+
(item["fileIdentifier"], hf_object_paths[uid], item["sha256"])
|
375 |
+
)
|
376 |
+
else:
|
377 |
+
versioned_dirname = os.path.join(download_dir, "hf-objaverse-v1")
|
378 |
+
fs, path = fsspec.core.url_to_fs(versioned_dirname)
|
379 |
+
|
380 |
+
# Get the existing file paths. This is much faster than calling fs.exists() for each
|
381 |
+
# file. `glob()` is like walk, but returns a list of files instead of the nested
|
382 |
+
# directory structure. glob() is also faster than find() / walk() since it doesn't
|
383 |
+
# need to traverse the entire directory structure.
|
384 |
+
existing_file_paths = fs.glob(
|
385 |
+
os.path.join(path, "glbs", "*", "*.glb"), refresh=True
|
386 |
+
)
|
387 |
+
existing_uids = {
|
388 |
+
file.split("/")[-1].split(".")[0]
|
389 |
+
for file in existing_file_paths
|
390 |
+
if file.endswith(".glb") # note partial files end with .glb.tmp
|
391 |
+
}
|
392 |
+
|
393 |
+
# add the existing downloaded uids to the return dict
|
394 |
+
already_downloaded_uids = uids_set.intersection(existing_uids)
|
395 |
+
for uid in already_downloaded_uids:
|
396 |
+
hf_object_path = hf_object_paths[uid]
|
397 |
+
fs_abs_object_path = os.path.join(versioned_dirname, hf_object_path)
|
398 |
+
out[self.uid_to_file_identifier(uid)] = fs_abs_object_path
|
399 |
+
|
400 |
+
logger.info(
|
401 |
+
f"Found {len(already_downloaded_uids)} objects already downloaded"
|
402 |
+
)
|
403 |
|
404 |
+
# get the uids that need to be downloaded
|
405 |
+
remaining_uids = uids_set - existing_uids
|
406 |
+
for uid in remaining_uids:
|
407 |
+
item = objects_uid_index.loc[uid]
|
408 |
+
if uid not in hf_object_paths:
|
409 |
+
logger.error(f"Could not find object with uid {uid}. Skipping it.")
|
410 |
+
if handle_missing_object is not None:
|
411 |
+
handle_missing_object(
|
412 |
+
file_identifier=item["fileIdentifier"],
|
413 |
+
sha256=item["sha256"],
|
414 |
+
metadata=dict(),
|
415 |
+
)
|
416 |
+
continue
|
417 |
+
objects_to_download.append(
|
418 |
+
(item["fileIdentifier"], hf_object_paths[uid], item["sha256"])
|
419 |
+
)
|
420 |
|
421 |
+
logger.info(f"Downloading {len(objects_to_download)} new objects")
|
|
|
|
|
|
|
|
|
|
|
|
|
422 |
|
423 |
+
# check if all objects are already downloaded
|
424 |
+
if len(objects_to_download) == 0:
|
425 |
+
return out
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
426 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
427 |
args = [
|
428 |
+
(
|
429 |
+
file_identifier,
|
430 |
+
hf_object_path,
|
431 |
+
download_dir,
|
432 |
+
sha256,
|
433 |
+
handle_found_object,
|
434 |
+
handle_modified_object,
|
435 |
+
handle_missing_object,
|
436 |
+
)
|
437 |
+
for file_identifier, hf_object_path, sha256 in objects_to_download
|
438 |
]
|
439 |
|
440 |
# download the objects in parallel
|
441 |
+
with Pool(processes) as pool:
|
442 |
new_object_downloads = list(
|
443 |
tqdm(
|
444 |
+
pool.imap_unordered(self._parallel_download_object, args),
|
445 |
total=len(args),
|
446 |
)
|
447 |
)
|
448 |
|
449 |
+
for file_identifier, local_path in new_object_downloads:
|
450 |
+
out[file_identifier] = local_path
|
|
|
|
|
451 |
|
452 |
+
return out
|
453 |
|
454 |
+
def load_lvis_annotations(
|
455 |
+
self,
|
456 |
+
download_dir: str = "~/.objaverse",
|
457 |
+
) -> Dict[str, List[str]]:
|
458 |
+
"""Load the LVIS annotations.
|
459 |
|
460 |
+
If the annotations are not already downloaded, they will be downloaded.
|
461 |
|
462 |
+
Args:
|
463 |
+
download_dir: The base directory to download the annotations to. Supports all
|
464 |
+
file systems supported by fsspec. Defaults to "~/.objaverse".
|
465 |
|
466 |
+
Returns:
|
467 |
+
A dictionary mapping the LVIS category to the list of uids in that category.
|
468 |
+
"""
|
469 |
+
hf_url = "https://huggingface.co/datasets/allenai/objaverse/resolve/main/lvis-annotations.json.gz"
|
470 |
|
471 |
+
download_path = os.path.join(
|
472 |
+
download_dir, "hf-objaverse-v1", "lvis-annotations.json.gz"
|
473 |
+
)
|
474 |
|
475 |
+
# use fsspec
|
476 |
+
fs, path = fsspec.core.url_to_fs(download_path)
|
477 |
+
if not fs.exists(path):
|
478 |
+
# make dir if it doesn't exist
|
479 |
+
fs.makedirs(os.path.dirname(path), exist_ok=True)
|
480 |
|
481 |
+
# download the file
|
482 |
+
with fs.open(path, "wb") as f:
|
483 |
+
with urllib.request.urlopen(hf_url) as response:
|
484 |
+
f.write(response.read())
|
485 |
|
486 |
+
# load the gzip file
|
487 |
+
with fs.open(path, "rb") as f:
|
488 |
+
with gzip.GzipFile(fileobj=f) as gfile:
|
489 |
+
content = gfile.read()
|
490 |
+
data = json.loads(content)
|
491 |
|
492 |
+
return data
|