def getDimensionsFromModelId(modelId, modelInfoFilename, modelCatFilename):
modelInfo = ModelInformation(modelInfoFilename)
modelCat = ModelCategoryMapping(modelCatFilename)
refModelDimensions = None
otherSimilarDimensions = []
refModelId = str(modelId)
refCategory = modelCat.getCoarseGrainedCategoryForModelId(refModelId)
for modelId in modelInfo.model_info.keys():
category = modelCat.getCoarseGrainedCategoryForModelId(modelId)
if refCategory == category:
info = modelInfo.getModelInfo(modelId)
# FIXME: handle the general case where for the front vector, do not ignore
# NOTE: SUNCG is using the Y-up coordinate system
frontVec = info['front']
if np.count_nonzero(frontVec) > 1 or not np.array_equal(
frontVec, [0, 0, 1]):
continue
width, height, depth = info['aligned_dims'] / 100.0 # cm to m
otherSimilarDimensions.append([width, height, depth])
if refModelId == modelId:
refModelDimensions = np.array([width, height, depth])
otherSimilarDimensions = np.array(otherSimilarDimensions)
logger.debug('Number of similar objects found in dataset: %d' %
(otherSimilarDimensions.shape[0]))
# Volume statistics (assume a gaussian distribution)
# XXX: use a more general histogram method to define the categories,
# rather than simply comparing the deviation to the mean
refVolume = np.prod(refModelDimensions)
otherVolumes = np.prod(otherSimilarDimensions, axis=-1)
mean = np.mean(otherVolumes)
std = np.std(otherVolumes)
# Compare the deviation to the mean
overallSizeTag = None
diff = refVolume - mean
for tag, threshold in DimensionTable.overallSizeTable:
if threshold >= 0.0:
if diff > threshold * std:
overallSizeTag = tag
else:
if diff < threshold * std:
overallSizeTag = tag
if overallSizeTag is None:
overallSizeTag = 'normal'
return overallSizeTag
def testGetModelInfo(self):
info = ModelInformation(
os.path.join(TEST_SUNCG_DATA_DIR, "metadata", "models.csv"))
_ = info.getModelInfo('261')
def testInit(self):
_ = ModelInformation(
os.path.join(TEST_SUNCG_DATA_DIR, "metadata", "models.csv"))