search

microsoft / onnxruntime

ONNX Runtime: cross-platform, high performance ML inferencing and training accelerator
https://onnxruntime.ai
MIT License
14.45k stars 2.9k forks source link

implement compute kernel for ImageDecoder #17593

Open liqunfu opened 1 year ago

liqunfu commented 1 year ago

Describe the feature request

ImageDecoder has been add to ONNX 1.15.0: https://github.com/onnx/onnx/pull/5294. Need to implement compute kernel once ONNX 1.15.0 is in ORT (https://github.com/microsoft/onnxruntime/pull/17125). This issue shall include re-enable test of ImageDecoder op at onnxruntime/test/testdata/onnx_backend_test_series_filters.jsonc

Describe scenario use case

onnx_test_runner shal pass ONNX test cases for the ImageDecoder op

jantonguirao commented 9 months ago

I'd like to ask for an update on this request. Is it committed to be added to 1.17 release?

snnn commented 7 months ago

No. To implement the operator, we need to integrate codecs libraries to ONNX Runtime. We may either use a platform independent solution, like opencv, for all platforms, or develop separated code for each platform. Both have pros and cons. Haven't made a decision yet.