Philip Meier

icon company @Quansight icon location Germany icon location Senior Software Engineer working on PyTorch / torchvision

Results 359 comments of Philip Meier

torchvision.io.read_image support processing EXIF information in JPEG file

Sure, go ahead @Zekrom-7780!

[DONOTMERGE] add CI tests for torch.compile'ing the transforms.v2 kernels

With `torch-2.2.0.dev20231121+cpu` and `torchvision` @ 893b4abdc0c9df36c241c58769810f69e35dab48: - `backend='eager', dynamic=True`: 259 failed - `backend='inductor', dynamic=True`: 307 failed - `backend='eager', dynamic=False`: 98 failed - `backend='inductor', dynamic=False`: 110 failed

[DONOTMERGE] add CI tests for torch.compile'ing the transforms.v2 kernels

Some relevant issues on core that will resolve a lot of the issues: - pytorch/pytorch#114218 - ~pytorch/pytorch#114220~ avoided by #8171 - pytorch/pytorch#114231 - pytorch/pytorch#114310 - pytorch/pytorch#114353 - pytorch/pytorch#114354 - pytorch/pytorch#114464...

[DONOTMERGE] add CI tests for torch.compile'ing the transforms.v2 kernels

Failures on `pad_mask` with `fullgraph` are a test error and fixed in #8132.

[DONOTMERGE] add CI tests for torch.compile'ing the transforms.v2 kernels

In c5c72abeef04db4852e7ffb764c1c872aacbd25c and 29ea48a71df246bbcc9a4b4dff44baeaa45513fe, I've added `atol=1, rtol=0` for uint8 / bilinear resize. With this, the most lenient setting, i.e. static shapes, eager backend, and graphbreaks being allowed, the tests...

[DONOTMERGE] add CI tests for torch.compile'ing the transforms.v2 kernels

With `torch-2.3.0.dev20231218+cpu` and torchvision @ 6c2e0ae88b056ba2ac897d4a7c1b7153cefcb444: | dynamic | backend | fullgraph | failing tests | | --- | --- | --- | --: | | False | eager |...

[DONOTMERGE] add CI tests for torch.compile'ing the transforms.v2 kernels

I've factored out e6a54bf78732a9c27f2dfb56364219d726a7a4bd into #8171.

[DONOTMERGE] add CI tests for torch.compile'ing the transforms.v2 kernels

With `torch-2.3.0.dev20231222+cpu` and torchvision @ 26fb5efe9d046ed9bf059b393f5785b9bdd9ec7e: | dynamic | backend | fullgraph | failing tests | diff to [previous](https://github.com/pytorch/vision/pull/8127#issuecomment-1860790124) | --- | --- | --- | --: | --: |...

Added a flag 'process_exif: bool = False' to 'read_image'

> I didn't know if Torch had any way to read the image's EXIF information, so I used PIL to do that Torch does not have such functionality yet. However,...

Added a flag 'process_exif: bool = False' to 'read_image'

> So, sould i completely erase the changes that I made here? Mostly yes I'm afraid. To avoid this in the future, please talk to us to get guidance if...