NVIDIA / DALI

A GPU-accelerated library containing highly optimized building blocks and an execution engine for data processing to accelerate deep learning training and inference applications.
https://docs.nvidia.com/deeplearning/dali/user-guide/docs/index.html
Apache License 2.0
5.07k stars 615 forks source link

Add support for AVIF images #5613

Open clementpoiret opened 2 weeks ago

clementpoiret commented 2 weeks ago

Is this a new feature, an improvement, or a change to existing functionality?

New Feature

How would you describe the priority of this feature request

Should have (e.g. Adoption is possible, but the performance shortcomings make the solution inferior).

Please provide a clear description of problem this feature solves

Storing a huge amount of images to train DNN is a critical point. Storing those images in a modern format with interesting compression properties such as AVIF allows to reduce storage space and network transfer times.

Feature Description

As a research engineer, I want to be able to use the avif format in order to diminish my training costs and make my R&D pipeline faster.

Describe your ideal solution

A way to load AVIF images in the same way I can load jpeg, png or wepb images.

Describe any alternatives you have considered

I can load the images with pillow, but it would be more easier / better to do everything with dali.

Additional context

No response

Check for duplicates

jantonguirao commented 2 weeks ago

@clementpoiret Thank you for your request. We will consider it and let you know once we have any particular plans. Note: The feature in any case would be incorporated to nvImageCodec library (https://github.com/NVIDIA/nvImageCodec/) first, which DALI uses as a dependency

clementpoiret commented 2 weeks ago

Great, thanks for your update!