Add RESIZE operators to Realy TFLite frontend#3370
Conversation
|
@FrozenGene @yongwww Could you have a look? |
python/tvm/relay/frontend/tflite.py
Outdated
| align_corners = resize_options.AlignCorners() | ||
|
|
||
| # Use layout NHWC | ||
| out = _op.image.resize(in_expr, size, "NHWC", method, align_corners) |
There was a problem hiding this comment.
After previous, we could use tuple(target_size)
python/tvm/relay/frontend/tflite.py
Outdated
| # size - 1-D int32 Tensor of 2 elements: new_height, new_width | ||
| size = self.get_tensor_value(input_tensors[1]) | ||
| # convert numpy array to tuple | ||
| size = (size[0], size[1]) |
There was a problem hiding this comment.
suggest better name. Like this:
target_size = self.get_tensor_value(input_tensors[1])|
Frontend GPU test group failed It means that Jenkins worker still uses old tflite flatbuffers schema from Tensorflow 1.12 Schema v 1.13https://raw.githubusercontent.com/tensorflow/tensorflow/r1.13/tensorflow/lite/schema/schema.fbs |
|
I opened an issue for the build error #3373 |
|
Fixed the PR to support both tflite versions - 1.12 and 1.13 |
|
@kevinthesun Can you merge it? |
Add RESIZE operators to Realy TFLite frontend
Request to support
RESIZE_BILLINEARoperators comes from #3189