gemini-code-assist[bot] commented on code in PR #19864:
URL: https://github.com/apache/tvm/pull/19864#discussion_r3450289706
##########
python/tvm/relax/transform/legalize_ops/image.py:
##########
@@ -66,6 +66,7 @@ def _image_affine_grid(bb: BlockBuilder, call: Call) -> Expr:
topi.image.affine_grid,
call.args[0],
target_shape=target_shape,
+ align_corners=bool(call.attrs.align_corners),
Review Comment:
The `bool()` conversion is redundant here because `call.attrs.align_corners`
is already a boolean value (as defined in `AffineGridAttrs`). We can simplify
this to directly use `call.attrs.align_corners`.
```suggestion
align_corners=call.attrs.align_corners,
```
##########
python/tvm/topi/image/grid_sample.py:
##########
@@ -47,13 +51,22 @@ def affine_grid(data, target_shape):
)
dtype = data.dtype
- y_step = tirx.const((2.0 - 1e-7) / (target_shape[0] - 1), dtype=dtype)
- x_step = tirx.const((2.0 - 1e-7) / (target_shape[1] - 1), dtype=dtype)
- start = tirx.const(-1.0, dtype=dtype)
+ height, width = target_shape[0], target_shape[1]
+ if align_corners:
Review Comment:
When `align_corners` is `False`, the target shape dimensions are allowed to
be `1` because the pixel center mapping formula `(2 * i + 1) / size - 1` does
not result in a division by zero (unlike the `align_corners=True` formula which
divides by `size - 1`). However, the assertion on line 49 (`assert
target_shape[0] > 1 and target_shape[1] > 1`) unconditionally restricts both
dimensions to be greater than 1. Consider relaxing that assertion to only apply
when `align_corners` is `True`.
--
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]
