ALinrunrun opened a new issue, #19500:
URL: https://github.com/apache/tvm/issues/19500
### Expected behavior
TVM Relay should import and execute ONNX `Resize` consistently with ONNX
Runtime for:
```
mode="nearest"
coordinate_transformation_mode="half_pixel"
nearest_mode="round_prefer_floor"
```
For a 3x3 input upsampled by scales `[1.0, 1.0, 1.5, 1.5`], ONNX Runtime
produces:
```
[[0. 0. 1. 2.]
[0. 0. 1. 2.]
[3. 3. 4. 5.]
[6. 6. 7. 8.]]
```
### Actual behavior
TVM Relay produces a different output from ONNX Runtime:
```
TVM output [0,0]:
[[0. 1. 1. 2.]
[3. 4. 4. 5.]
[3. 4. 4. 5.]
[6. 7. 7. 8.]]
max_abs=4.000000e+00
```
The discrepancy appears when importing ONNX Resize with half_pixel and
round_prefer_floor.
### Environment
- TVM: 0.14.dev273 / Relay ONNX frontend
- ONNX Runtime: 1.23
- ONNX: 1.15.0
- NumPy: 1.26.4
- Python: 3.11
- Target: llvm
- OS: Linux
### Steps to reproduce
```
import numpy as np
import onnx
from onnx import TensorProto, helper
import onnxruntime as ort
import tvm
from tvm import relay
from tvm.contrib import graph_executor
def build_model():
x = helper.make_tensor_value_info("X", TensorProto.FLOAT, [1, 1, 3, 3])
y = helper.make_tensor_value_info("Y", TensorProto.FLOAT, None)
scales = helper.make_tensor(
"scales", TensorProto.FLOAT, [4], [1.0, 1.0, 1.5, 1.5]
)
resize = helper.make_node(
"Resize",
["X", "", "scales"],
["Y"],
mode="nearest",
coordinate_transformation_mode="half_pixel",
nearest_mode="round_prefer_floor",
)
graph = helper.make_graph([resize], "g", [x], [y], initializer=[scales])
return helper.make_model(graph, opset_imports=[helper.make_opsetid("",
13)])
x = np.arange(9, dtype=np.float32).reshape(1, 1, 3, 3)
model = build_model()
ort_sess = ort.InferenceSession(
model.SerializeToString(), providers=["CPUExecutionProvider"]
)
ort_out = ort_sess.run(None, {"X": x})[0]
mod, params = relay.frontend.from_onnx(model, shape={"X": list(x.shape)})
with tvm.transform.PassContext(opt_level=3):
lib = relay.build(mod, target="llvm", params=params)
gm = graph_executor.GraphModule(lib["default"](tvm.cpu()))
gm.set_input("X", x)
gm.run()
tvm_out = gm.get_output(0).numpy()
print("ORT output:")
print(ort_out[0, 0])
print("TVM output:")
print(tvm_out[0, 0])
print("max_abs:", np.max(np.abs(ort_out - tvm_out)))
```
### Triage
needs-triage
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
