def loop_fn(i):
grad1 = array_ops.gather(grad, i)
return nn.conv2d_backprop_input(x_shape,
filt,
grad1,
strides=[1, 2, 2, 1],
padding="VALID",
data_format="NHWC")
def loop_fn(i):
grad1 = array_ops.gather(grad, i)
return nn.conv2d_backprop_input(
x_shape,
filt,
grad1,
strides=[1, 2, 2, 1],
padding="VALID",
data_format="NHWC")
def template(N=2,
Hin=5,
Hout=3,
Win=5,
Wout=3,
Cin=7,
Cout=9,
KH=3,
KW=3,
strides=[1, 1, 1, 1],
padding="VALID",
data_format="NHWC",
description: str = ""):
from tensorflow.python.ops import nn
x_shape = [N, Hin, Win, Cin
] if data_format == "NHWC" else [N, Cin, Hin, Win]
y_shape = [N, Hout, Wout, Cout
] if data_format == "NHWC" else [N, Cout, Hout, Wout]
w_shape = [KH, KW, Cin, Cout]
w = tf.placeholder(np.float32, w_shape)
gy = tf.placeholder(np.float32, y_shape)
x = nn.conv2d_backprop_input(x_shape,
w,
gy,
strides=strides,
padding=padding,
data_format=data_format)
vw = np.random.rand(*w_shape).astype(np.float32)
vgy = np.random.rand(*y_shape).astype(np.float32)
with tf.Session() as sess:
vx, = sess.run([x], {w: vw, gy: vgy})
graph = TensorFlowConverter(sess, batch_size=2).convert([w, gy], [x])
generate_kernel_test_case(
description=f"[TensorFlow] Conv2DBackPropInput {description}",
graph=graph,
backend=["webgpu", "webassembly", "webgl"],
inputs={
graph.inputs[0]: vw,
graph.inputs[1]: vgy
},
expected={graph.outputs[0]: vx})
