add image resize operator and support batch_size #13611
Conversation
|
@zhreshold @apeforest - Can you help take a review? |
|
@mxnet-label-bot add[Operator, pr-awaiting-review] |
stu1130
changed the title
| DMLC_DECLARE_FIELD(size) | ||
| .set_default(nnvm::Tuple<int>()) | ||
| .describe("Size of new image. Could be (width, height) or (size)"); | ||
| DMLC_DECLARE_FIELD(keep_ratio) |
There was a problem hiding this comment.
why not setting set_default for keep_ratio ?
There was a problem hiding this comment.
python fronted will have a default value for keep_ratio
but no hurt to set default value here
src/operator/image/resize.cc
Outdated
| DMLC_REGISTER_PARAMETER(ResizeParam); | ||
|
|
||
| NNVM_REGISTER_OP(_image_resize) | ||
| .describe(R"code()code" ADD_FILELINE) |
There was a problem hiding this comment.
can we add some usage examples.
| data_expected = image.imresize(data_in, 200, 200, 1) | ||
| assert_almost_equal(out_nd.asnumpy(), data_expected.asnumpy()) | ||
| # test 4D input | ||
| data_bath_in = nd.random.uniform(0, 255, (3, 300, 200, 3)).astype('uint8') |
There was a problem hiding this comment.
can we test for one other dtype apart from uint8
| MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, { | ||
| cv::Mat buf(inputs[0].shape_[kH], inputs[0].shape_[kW], cv_type, | ||
| inputs[0].dptr<DType>() + input_index); | ||
| cv::Mat dst(outputs[0].shape_[kH], outputs[0].shape_[kW], cv_type, |
There was a problem hiding this comment.
are these calls thread safe ?
There was a problem hiding this comment.
for each iteration, we access different address of the input and write the result back to the different address as well. It's thread safe
|
@stu1130 Can you look into failing CI builds? Thanks |
|
Close the PR to implement GPU support, will reopen once it's done |
Description
Checklist
Essentials
Please feel free to remove inapplicable items for your PR.
Changes
Comments
@sandeep-krishnamurthy