Is DALI any good anymore?

[BrechtCorbeel]

Describe the question.

I have not been keeping up, it's been years since use, but how has it improved and how does DALI stack up against tensor models today?

Check for duplicates

• I have searched the open bugs/issues and have found no duplicates for this bug report

[JanuszL]

Hi @BrechtCorbeel,

Thank you for reaching out.

In short, yes, DALI is still applicable. If it's been a while since you last tried it, you'll be pleased to know that we've made significant improvements in recent releases:

• Improved execution flow for better memory utilization (up to 40% in some cases) and the ability to move data CPU->GPU->CPU->GPU within a single pipeline.
• DALI proxy allows you to integrate DALI into existing data processing workflows in PyTorch without needing to convert the entire data processing definition, just the relevant parts.
• Checkpointing the state of the DALI pipeline.
• And many more improvements, which you can find in detail here.
  If you have any specific use cases you need to accelerate, please share them here so we can explore how DALI can help.