fundamentalvision / Deformable-DETR

Deformable DETR: Deformable Transformers for End-to-End Object Detection.
Apache License 2.0
3.22k stars 520 forks source link

Dataset and Hardware Requirements for fine-tuning and quick inference. #153

Open linsecDev opened 2 years ago

linsecDev commented 2 years ago

Hi @jackroos, Thank you for your valuable work. I have two follow-up questions. Will highly appreciate your response.

1- What should be the preferable size of the custom dataset for fine-tuning to get good results?

2- On which cloud platform (AWS, Google cloud, etc.) should I set-up pre-trained DETR for quick inference. Can you hint on some preferable hardware resources (GPU, CPU, Memory) required for fine-tuning and quick inference. I've been researching on this lately, but couldn't get some reasonable answer specifically for DETR.

Will highly appreciate response. Thanks!