MCPcopy Create free account
hub / github.com/QWTforGithub/T2LDM / preprocess

Function preprocess

utils/common.py:228–295  ·  view source on GitHub ↗
(
        batch,
        classifier_dropout=0.1,
        use_text=False,
        use_semantic=False,
        train_depth=True,
        train_reflectance=True,
        lidar_utils=None,
        text_name="text_aim"
)

Source from the content-addressed store, hash-verified

226 return colors
227
228def preprocess(
229 batch,
230 classifier_dropout=0.1,
231 use_text=False,
232 use_semantic=False,
233 train_depth=True,
234 train_reflectance=True,
235 lidar_utils=None,
236 text_name="text_aim"
237):
238 x = []
239 if train_depth:
240 x += [lidar_utils.convert_depth(batch["depth"])]
241 if train_reflectance:
242 x += [batch["reflectance"]]
243 x = torch.cat(x, dim=1)
244 x = lidar_utils.normalize(x)
245 x = F.interpolate(
246 x.to("cuda"),
247 size=lidar_utils.resolution,
248 mode="nearest-exact",
249 )
250
251 new_texts = None
252 texts = None
253 if (use_text):
254 texts = batch[text_name]
255 new_texts = []
256 for text in texts:
257 if random.random() < classifier_dropout:
258 new_texts.append("")
259 else:
260 new_texts.append(text)
261
262 semantic = None
263 if (use_semantic):
264 semantic = batch["semantic"]
265 if(not semantic.is_cuda):
266 semantic = semantic.cuda()
267
268 xyz = None
269 if("xyz" in batch.keys()):
270 xyz = batch["xyz"]
271 if(not xyz.is_cuda):
272 xyz = xyz.cuda()
273
274 points = None
275 if("points" in batch.keys()):
276 points = batch["points"]
277 if(not points.is_cuda):
278 points = points.cuda()
279
280 batches = None
281 if("batch" in batch.keys()):
282 batches = batch["batch"]
283 if(not batches.is_cuda):
284 batches = batches.cuda()
285

Callers

nothing calls this directly

Calls 3

convert_depthMethod · 0.80
normalizeMethod · 0.80
interpolateMethod · 0.80

Tested by

no test coverage detected