MLBEDSW-7542: Fix output diff caused by wrong scaling in Conv2d
- The reference calculates the scale slightly different between
Conv2d and FullyConnect. Recently a fix was submitted to address
this issue. However, internally a Conv2d can be changed to a
FullyConnect but then the scale must still be calculated
following the Conv2d reference.
- The fix is to check the original type if FullyConnect scale
should be used or not.
Change-Id: I5a9fb49126f0df63712b73fb5520fdc604cee378
Signed-off-by: Johan Alfven <johan.alfven@arm.com>
diff --git a/ethosu/vela/weight_compressor.py b/ethosu/vela/weight_compressor.py
index a37ff6a..a258be4 100644
--- a/ethosu/vela/weight_compressor.py
+++ b/ethosu/vela/weight_compressor.py
@@ -284,7 +284,7 @@
# TensorFlow Lite casts the scales slightly differently for uint8 and int8 as well as
# for FullyConnected operators
if not rescale_for_faf:
- if ifm_dtype == DataType.uint8 or first_consumer_op.type == Op.FullyConnected:
+ if ifm_dtype == DataType.uint8 or first_consumer_op.original_type == Op.FullyConnected:
scales = [np.double(ifm_scale * weight_scale) / np.double(ofm_scale) for weight_scale in weight_scales]
elif ifm_dtype == DataType.int8 or ifm_dtype == DataType.int16:
scales = [