face_landmark_with_attention to OAK

Author: PINTO0309

setup.py

@@ -11,7 +11,7 @@
     setup(
         name="tflite2tensorflow",
         scripts=scripts,
-        version="1.20.7",
+        version="1.20.8",
         description="Generate saved_model, tfjs, tf-trt, EdgeTPU, CoreML, quantized tflite, ONNX, OpenVINO, Myriad Inference Engine blob and .pb from .tflite.",
         long_description=long_description,
         long_description_content_type="text/markdown",

tflite2tensorflow/mediapipeCustomOp.py

@@ -72,7 +72,7 @@ def TransformLandmarks(operator, custom_options, tensors, interpreter, landmarks
     return landmarks2d_transformed
 
 #Affine transform images using bilinear interpolation
-def TransformTensorBilinear(operator, custom_options, tensors, interpreter, optimizing_barracuda, features=None, mat=None):
+def TransformTensorBilinear(operator, custom_options, tensors, interpreter, optimizing_barracuda, optimaization_for_myriad, features=None, mat=None):
     if features is None:
         features = tensors[operator['inputs'][0]] #float32 [b,48,48,32] feature maps
     if mat is None:
@@ -110,8 +110,8 @@ def TransformTensorBilinear(operator, custom_options, tensors, interpreter, opti
     weight_floor = tf.subtract(tf.ones(2), weight_ceil_) #[b,h,w,2]
     weight_ceilX = tf.multiply(weight_ceil_[:,:,:,0:1], weight_floor[:,:,:,1:2]) #[b,h,w]
     weight_ceilY = tf.multiply(weight_floor[:,:,:,0:1], weight_ceil_[:,:,:,1:2]) #[b,h,w]
-    weight_ceil_ = tf.reduce_prod(weight_ceil_, axis=3, keepdims=True) #[b,h,w,1]
-    weight_floor = tf.reduce_prod(weight_floor, axis=3, keepdims=True) #[b,h,w,1]
+    weight_ceil_ = tf.multiply(weight_ceil_[:,:,:,0:1], weight_ceil_[:,:,:,1:2]) #[b,h,w]
+    weight_floor = tf.multiply(weight_floor[:,:,:,0:1], weight_floor[:,:,:,1:2]) #[b,h,w]
 
     # Find nearest 4 points.
     # Make sure they are in the input image
@@ -157,7 +157,7 @@ def barracuda_gather_nd(params, indices):
         return tf.expand_dims(tf.reshape(result_flat, tf.concat([idx_shape[:-1], gather_shape], axis=0)), axis=0)
 
     # calc final pixel value
-    if not optimizing_barracuda:
+    if not optimizing_barracuda and not optimaization_for_myriad:
         value_floor = tf.gather_nd(params=features, indices=in_coord_floor, batch_dims=1) #[b,h,w,32]
         value_ceilX = tf.gather_nd(params=features, indices=in_coord_ceilX, batch_dims=1) #[b,h,w,32]
         value_ceilY = tf.gather_nd(params=features, indices=in_coord_ceilY, batch_dims=1) #[b,h,w,32]

tflite2tensorflow/tflite2tensorflow.py

@@ -3485,7 +3485,8 @@ def barracuda_gather_nd(params, indices):
                 result_flat = tf.gather(params_flat, indices_flat)
                 return tf.expand_dims(tf.reshape(result_flat, tf.concat([idx_shape[:-1], gather_shape], axis=0)), axis=0)
 
-            if not optimizing_barracuda:
+            optimaization_for_myriad = (optimizing_for_openvino_and_myriad and rigorous_optimization_for_myriad)
+            if not optimizing_barracuda and not optimaization_for_myriad:
                 output_tensor = tf.gather_nd(
                     input_tensor1,
                     input_tensor2,
@@ -5126,7 +5127,15 @@ def complexabs_(x, tout):
                     options = op['custom_options']
                     custom_options = read_flexbuffer(np.array(options, dtype=np.uint8).tobytes())
                     output_detail = interpreter._get_tensor_details(op['outputs'][0])
-                    tensors[output_detail['index']] = TransformTensorBilinear(op, custom_options, tensors, interpreter, optimizing_barracuda)
+                    optimaization_for_myriad = (optimizing_for_openvino_and_myriad and rigorous_optimization_for_myriad)
+                    tensors[output_detail['index']] = TransformTensorBilinear(
+                        op,
+                        custom_options,
+                        tensors,
+                        interpreter,
+                        optimizing_barracuda,
+                        optimaization_for_myriad,
+                    )
 
                 elif custom_op_type == 'TransformLandmarks':
                     custom_options = None