administrator/frigate
1
0
Fork 0
mirror of https://github.com/blakeblackshear/frigate.git synced 2025-12-06 13:34:13 +03:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

fix formatting

Browse Source
This commit is contained in:
Dan Brown 2025-12-02 21:02:01 +01:00
parent 6ee36d6ffc
commit f5f84a3c53
1 changed files with 6 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
frigate/detectors/plugins/edgetpu_tfl.py
View File

@ -114,9 +114,7 @@ class EdgeTpuTfl(DetectionApi):
# to differentiate from (not used) max score tensor # to differentiate from (not used) max score tensor
output_classes_index = i output_classes_index = i
if output_boxes_index is None or output_classes_index is None: if output_boxes_index is None or output_classes_index is None:
logger.warning( logger.warning("Unrecognized model output, unexpected tensor shapes.")
"Unrecognized model output, unexpected tensor shapes."
)
output_classes_index = ( output_classes_index = (
0 0
if (output_boxes_index is None or output_classes_index == 1) if (output_boxes_index is None or output_classes_index == 1)
@ -127,17 +125,13 @@ class EdgeTpuTfl(DetectionApi):
scores_details = self.tensor_output_details[output_classes_index] scores_details = self.tensor_output_details[output_classes_index]
classes_count = scores_details["shape"][2] classes_count = scores_details["shape"][2]
self.scores_tensor_index = scores_details["index"] self.scores_tensor_index = scores_details["index"]
self.scores_scale, self.scores_zero_point = scores_details[ self.scores_scale, self.scores_zero_point = scores_details["quantization"]
"quantization"
]
# calculate the quantized version of the min_score # calculate the quantized version of the min_score
self.min_score_quantized = int( self.min_score_quantized = int(
(self.min_logit_value / self.scores_scale) + self.scores_zero_point (self.min_logit_value / self.scores_scale) + self.scores_zero_point
) )
self.logit_shift_to_positive_values = ( self.logit_shift_to_positive_values = (
max( max(0, math.ceil((128 + self.scores_zero_point) * self.scores_scale))
0, math.ceil((128 + self.scores_zero_point) * self.scores_scale)
)
+ 1 + 1
) # round up ) # round up
@ -245,9 +239,7 @@ class EdgeTpuTfl(DetectionApi):
scores_output_quantized = self.interpreter.get_tensor( scores_output_quantized = self.interpreter.get_tensor(
self.scores_tensor_index self.scores_tensor_index
)[0] # (2100, NC) )[0] # (2100, NC)
max_scores_quantized = np.max( max_scores_quantized = np.max(scores_output_quantized, axis=1) # (2100,)
scores_output_quantized, axis=1
) # (2100,)
mask = max_scores_quantized >= self.min_score_quantized # (2100,) mask = max_scores_quantized >= self.min_score_quantized # (2100,)
if not np.any(mask): if not np.any(mask):
@ -277,9 +269,7 @@ class EdgeTpuTfl(DetectionApi):
# Softmax over the 16 bins # Softmax over the 16 bins
dfl_max = np.max(dfl_distributions, axis=2, keepdims=True) dfl_max = np.max(dfl_distributions, axis=2, keepdims=True)
dfl_exp = np.exp(dfl_distributions - dfl_max) dfl_exp = np.exp(dfl_distributions - dfl_max)
dfl_probs = dfl_exp / np.sum( dfl_probs = dfl_exp / np.sum(dfl_exp, axis=2, keepdims=True) # (N, 4, 16)
dfl_exp, axis=2, keepdims=True
) # (N, 4, 16)
# Weighted sum: (N, 4, 16) * (16,) -> (N, 4) # Weighted sum: (N, 4, 16) * (16,) -> (N, 4)
distances = np.einsum("pcr,r->pc", dfl_probs, self.project) distances = np.einsum("pcr,r->pc", dfl_probs, self.project)